Lianmin Zheng is a hands-on systems and ML engineer (Prev: xAI) with a Ph.D. from UC Berkeley and co-founder of lmsys.org. He builds high-performance model runtimes and kernels, shipping practical optimizations that make cutting-edge models run faster and cheaper. His X feed blends deep technical dives with career-making engineering wins.
You optimize kernels so obsessively that if someone asked you to reduce the social latency of a dinner party you’d benchmark forks and compile the seating arrangement into PTX, then tweet a three-part thread about it.
Led the engineering work that enabled new xAI image and autoregressive models to run efficiently (including the zero-CPU-overhead scheduler) and helped ship performance that surprised the field, contributing to headlines like Grok breaking 1400 Elo.
To push the boundaries of practical ML performance, making large models and novel architectures accessible by squeezing out every inefficiency in software and hardware, and sharing the tools and knowledge to let others do the same.
Technical rigor and reproducibility matter more than hype; well-engineered systems unlock research impact; open collaboration accelerates progress; small, well-targeted optimizations compound into massive practical gains.
Deep low-level expertise (kernels, compilers, schedulers), proven track record of shipping high-impact optimizations, credibility from elite research and industry roles, and the ability to explain technical wins in short, resonant posts.
Tends to live in niche, deeply technical territory, great for peers but sometimes opaque for broader audiences; can prioritize micro-optimizations over broader messaging or product storytelling.
Post regular explainer threads that bridge deep tech and practical takeaways: short first tweet with the headline result, follow-up tweets with a simple diagram, a code snippet or benchmark, and a clear call to try or reproduce. Pair that with demo notebooks, short video clips of profiling/benchmarks, and occasional high-level threads that translate kernel wins into cost/time/user benefits. Engage in AMAs, retweet community reproductions, and tag relevant projects to turn niche expertise into broader influence on X.
Fun fact: while optimizing SGLang inference, Lianmin spent more time eliminating CPU overhead than tuning GPU kernels, he built a zero-CPU-overhead batch scheduler. He’s previously worked at xAI, holds a Ph.D. from UC Berkeley, cofounded lmsys.org, and has ~15.9k followers on X.
Grok-2 is here, a new frontier-level model from @xAI!
I still remember the good old days when I was a GPU-poor grad student, playing with the Vicuna model and building the Chatbot Arena leaderboard with just a few GPUs.
But now, my job at xAI is developing systems for the 100K GPU cluster. If you want to be among the first to see AGI, join xAI at x.ai/careers.
I’ve learned a ton from @ying11231 and @elonmusk this year. The way they think and the instincts they rely on when tackling challenges are insanely similar.
Only when you work closely with them do you start to realize just how visionary, consistent, and truth-seeking they really are.
Grok-2 is here, a new frontier-level model from @xAI!
I still remember the good old days when I was a GPU-poor grad student, playing with the Vicuna model and building the Chatbot Arena leaderboard with just a few GPUs.
But now, my job at xAI is developing systems for the 100K GPU cluster. If you want to be among the first to see AGI, join xAI at x.ai/careers.
At #NeurIPS2024 this week. Happy to chat about anything related to LLMs and xAI: our 100K training cluster, inference stack, Grok3, and the newly released AR image gen model.
We're presenting an SGLang poster on Thursday afternoon. Welcome to stop by and join the discussion:
neurips.cc/virtual/2024/p…
{"data":{"__meta":{"device":false,"path":"/creators/lm_zheng"},"/creators/lm_zheng":{"data":{"user":{"id":"952696452173594625","name":"Lianmin Zheng","description":"Prev: Engineer @xAI, Ph.D. @UCBerkeley, Co-founder @lmsysorg","followers_count":15933,"friends_count":658,"statuses_count":594,"profile_image_url_https":"https://pbs.twimg.com/profile_images/1107058309994024960/kUJgsrqM_normal.jpg","screen_name":"lm_zheng","location":"Bay Area, California","entities":{"description":{"urls":[]},"url":{"urls":[{"display_url":"lmzheng.net","expanded_url":"http://lmzheng.net/","indices":[0,23],"url":"https://t.co/c17mLv9RxT"}]}}},"details":{"type":"The Innovator","description":"Lianmin Zheng is a hands-on systems and ML engineer (Prev: xAI) with a Ph.D. from UC Berkeley and co-founder of lmsys.org. He builds high-performance model runtimes and kernels, shipping practical optimizations that make cutting-edge models run faster and cheaper. His X feed blends deep technical dives with career-making engineering wins.","purpose":"To push the boundaries of practical ML performance—making large models and novel architectures accessible by squeezing out every inefficiency in software and hardware, and sharing the tools and knowledge to let others do the same.","beliefs":"Technical rigor and reproducibility matter more than hype; well-engineered systems unlock research impact; open collaboration accelerates progress; small, well-targeted optimizations compound into massive practical gains.","facts":"Fun fact: while optimizing SGLang inference, Lianmin spent more time eliminating CPU overhead than tuning GPU kernels—he built a zero-CPU-overhead batch scheduler. He’s previously worked at xAI, holds a Ph.D. from UC Berkeley, cofounded lmsys.org, and has ~15.9k followers on X.","strength":"Deep low-level expertise (kernels, compilers, schedulers), proven track record of shipping high-impact optimizations, credibility from elite research and industry roles, and the ability to explain technical wins in short, resonant posts.","weakness":"Tends to live in niche, deeply technical territory—great for peers but sometimes opaque for broader audiences; can prioritize micro-optimizations over broader messaging or product storytelling.","roast":"You optimize kernels so obsessively that if someone asked you to reduce the social latency of a dinner party you’d benchmark forks and compile the seating arrangement into PTX—then tweet a three-part thread about it.","win":"Led the engineering work that enabled new xAI image and autoregressive models to run efficiently (including the zero-CPU-overhead scheduler) and helped ship performance that surprised the field—contributing to headlines like Grok breaking 1400 Elo.","recommendation":"Post regular explainer threads that bridge deep tech and practical takeaways: short first tweet with the headline result, follow-up tweets with a simple diagram, a code snippet or benchmark, and a clear call to try or reproduce. Pair that with demo notebooks, short video clips of profiling/benchmarks, and occasional high-level threads that translate kernel wins into cost/time/user benefits. Engage in AMAs, retweet community reproductions, and tag relevant projects to turn niche expertise into broader influence on X."},"tweets":[{"bookmarked":false,"display_text_range":[0,155],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[{"id_str":"1661523610111193088","name":"xAI","screen_name":"xai","indices":[135,139]}]},"favorited":false,"lang":"en","quoted_status_id_str":"1891706264800936307","quoted_status_permalink":{"url":"https://t.co/fhF5Mm3HQg","expanded":"https://twitter.com/arena/status/1891706264800936307","display":"x.com/arena/status/1…"},"retweeted":false,"fact_check":null,"id":"1891713559874187417","view_count":28678,"bookmark_count":25,"created_at":1739854633000,"favorite_count":699,"quote_count":6,"reply_count":21,"retweet_count":51,"user_id_str":"952696452173594625","conversation_id_str":"1891713559874187417","full_text":"People have been wondering which model will be the first to reach a 1400 Elo score; no one believed it would be Grok one year ago, but @xAI made it happen!","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[]},"display_text_range":[0,262],"lang":"en","quoted_tweet":{"id":"1990866667237191758","text":"Never block the GPU!\n\nIn a new @modal blogpost, we walk through a major class of inefficiency in AI inference: host overhead.\n\nWe include three cases where we worked with @sgl_project to cut host overhead and prevent GPU stalls.\n\nEvery microsecond counts.\n\nhttps://t.co/ZeumrZpSKE","full_text":"Never block the GPU!\n\nIn a new @modal blogpost, we walk through a major class of inefficiency in AI inference: host overhead.\n\nWe include three cases where we worked with @sgl_project to cut host overhead and prevent GPU stalls.\n\nEvery microsecond counts.\n\nhttps://t.co/ZeumrZpSKE","created_at":1763494576000,"author_id":"950189417754513409","author":{"id":"950189417754513409","name":"Charles 🎉 Frye","username":"charles_irl","screen_name":"charles_irl","profile_image_url":"https://pbs.twimg.com/profile_images/1320840184494108674/d7A64nIG_400x400.jpg","profile_image_url_https":"https://pbs.twimg.com/profile_images/1320840184494108674/d7A64nIG_400x400.jpg","is_blue_verified":1},"public_metrics":{"like_count":416,"retweet_count":40,"reply_count":6,"quote_count":10}},"fact_check":null,"id":"1992497243686080769","view_count":103176,"bookmark_count":277,"created_at":1763883336000,"favorite_count":676,"quote_count":2,"reply_count":21,"retweet_count":30,"user_id_str":"952696452173594625","conversation_id_str":"1992497243686080769","full_text":"Fun fact:\nWhile inference is running on the GPU, I probably spent more time optimizing CPU overhead than the GPU kernels themselves during my development of SGLang.\n\nMany big refactors and small details were needed to achieve a zero-CPU-overhead batch scheduler.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":1765371346955,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[{"id_str":"44196397","name":"Elon Musk","screen_name":"elonmusk","indices":[56,65]}]},"display_text_range":[0,114],"lang":"en","quoted_tweet":{"id":"2008683630768648493","text":"This is my favorite clip of the new Elon pod. He opens up saying xAI struggles with memory usage/bandwidth and CUDA kernel optimization (matmul, attention, MoE, etc). If you are good kernel or performance engineering in general, you should apply. Steer the world in a better direction.","full_text":"This is my favorite clip of the new Elon pod. He opens up saying xAI struggles with memory usage/bandwidth and CUDA kernel optimization (matmul, attention, MoE, etc). If you are good kernel or performance engineering in general, you should apply. Steer the world in a better direction.","created_at":1767742471000,"author_id":"1595935520999510016","author":{"id":"1595935520999510016","name":"Elliot Arledge","username":"elliotarledge","screen_name":"elliotarledge","profile_image_url":"https://pbs.twimg.com/profile_images/2005693506644574208/Hp8sqTPA_400x400.jpg","profile_image_url_https":"https://pbs.twimg.com/profile_images/2005693506644574208/Hp8sqTPA_400x400.jpg","is_blue_verified":1},"public_metrics":{"like_count":906,"retweet_count":30,"reply_count":34,"quote_count":14}},"fact_check":null,"id":"2009196655288217735","view_count":90962,"bookmark_count":104,"created_at":1767864786000,"favorite_count":538,"quote_count":4,"reply_count":35,"retweet_count":39,"user_id_str":"952696452173594625","conversation_id_str":"2009196655288217735","full_text":"Apply to the kernel team and you get the chance to show @elonmusk how to write efficient kernels in PTX for GB300.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"rapidapi","fetched_at":1770137356512,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"bookmarked":false,"display_text_range":[0,280],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[]},"favorited":false,"lang":"en","quoted_status_id_str":"1866224902711411050","quoted_status_permalink":{"url":"https://t.co/B16xxZBptv","expanded":"https://twitter.com/xai/status/1866224902711411050","display":"x.com/xai/status/186…"},"retweeted":false,"fact_check":null,"id":"1866280165254263115","view_count":132795,"bookmark_count":79,"created_at":1733790839000,"favorite_count":529,"quote_count":4,"reply_count":10,"retweet_count":27,"user_id_str":"952696452173594625","conversation_id_str":"1866280165254263115","full_text":"We released a new image generation model that was trained from scratch at xAI!\n\nIt has been fun optimizing the inference for it. Because it's an autoregressive model, we can easily port it to SGLang and reuse all the great optimizations like the new zero-overhead batch scheduler, torch.compile, and many custom kernels.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"bookmarked":false,"display_text_range":[0,200],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[]},"favorited":false,"lang":"en","quoted_status_id_str":"1872242657348710721","quoted_status_permalink":{"url":"https://t.co/jq2TrcpMiA","expanded":"https://twitter.com/deepseek_ai/status/1872242657348710721","display":"x.com/deepseek_ai/st…"},"retweeted":false,"fact_check":null,"id":"1872266844603904398","view_count":43801,"bookmark_count":55,"created_at":1735218175000,"favorite_count":515,"quote_count":5,"reply_count":18,"retweet_count":37,"user_id_str":"952696452173594625","conversation_id_str":"1872266844603904398","full_text":"Highly respected! It is so impressive given the results and the very limited resources they have compared to other big labs.\n\n\"DeepSeek-V3 is trained on a cluster equipped with 2048 NVIDIA H800 GPUs.\"","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"bookmarked":false,"display_text_range":[0,31],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[{"display_url":"x.com/i/events/19427…","expanded_url":"https://x.com/i/events/1942716886258528256","url":"https://t.co/Ody4Uwh1n2","indices":[8,31]}],"user_mentions":[]},"favorited":false,"lang":"en","possibly_sensitive":false,"possibly_sensitive_editable":true,"retweeted":false,"fact_check":null,"id":"1943153321801633805","view_count":84502,"bookmark_count":39,"created_at":1752118828000,"favorite_count":501,"quote_count":5,"reply_count":68,"retweet_count":38,"user_id_str":"952696452173594625","conversation_id_str":"1943153321801633805","full_text":"Grok4 🚀\nhttps://t.co/Ody4Uwh1n2","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":0,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[]},"display_text_range":[0,280],"lang":"en","quoted_tweet":{"id":"2011869695482249680","text":"Introducing SGLang Pipeline Parallelism: The highest-performance open-source PP implementation for 1M+ token contexts.\n\nWe’ve engineered a robust multi-node scaling solution that redefines efficiency: \n✨ Dynamic Chunking: Intelligently minimizes pipeline bubbles. \n⚡ Async P2P Communication: Maximize hardware utilization. \n🧩 Seamless Compatibility: Works perfectly with PD Disaggregation, HiCache, and hybrid parallelism.\n\nAchieve 3.31x throughput gains (vs TP8) and unlock the full potential of your trillion-parameter models. 👇\n\n#SGLang #LLM #DeepSeek #AIInference #OpenSource","full_text":"Introducing SGLang Pipeline Parallelism: The highest-performance open-source PP implementation for 1M+ token contexts.\n\nWe’ve engineered a robust multi-node scaling solution that redefines efficiency: \n✨ Dynamic Chunking: Intelligently minimizes pipeline bubbles. \n⚡ Async P2P Communication: Maximize hardware utilization. \n🧩 Seamless Compatibility: Works perfectly with PD Disaggregation, HiCache, and hybrid parallelism.\n\nAchieve 3.31x throughput gains (vs TP8) and unlock the full potential of your trillion-parameter models. 👇\n\n#SGLang #LLM #DeepSeek #AIInference #OpenSource","created_at":1768502088000,"author_id":"1822588444046249984","author":{"id":"1822588444046249984","name":"LMSYS Org","username":"lmsysorg","screen_name":"lmsysorg","profile_image_url":"https://pbs.twimg.com/profile_images/1823251762243121152/0cgIY6yc_400x400.jpg","profile_image_url_https":"https://pbs.twimg.com/profile_images/1823251762243121152/0cgIY6yc_400x400.jpg","is_blue_verified":1},"public_metrics":{"like_count":176,"retweet_count":23,"reply_count":3,"quote_count":8}},"fact_check":null,"id":"2012069688122003890","view_count":35150,"bookmark_count":206,"created_at":1768549770000,"favorite_count":364,"quote_count":2,"reply_count":9,"retweet_count":34,"user_id_str":"952696452173594625","conversation_id_str":"2012069688122003890","full_text":"Chunked pipeline parallelism is arguably the most general and scalable system technique for accelerating super-long-context inference.\n\nIt remains underrated today, largely because there still isn’t a strong, high-quality open-source implementation.\nThe SGLang team recently fully optimized it and published a detailed blog post explaining all the key details.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"rapidapi","fetched_at":1770776994668,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"bookmarked":false,"display_text_range":[0,279],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[{"display_url":"x.ai/careers","expanded_url":"https://x.ai/careers","url":"https://t.co/AlbqAelHiD","indices":[351,374]},{"display_url":"x.ai/careers","expanded_url":"https://x.ai/careers","url":"https://t.co/AlbqAelHiD","indices":[351,374]}],"user_mentions":[{"id_str":"1661523610111193088","name":"xAI","screen_name":"xai","indices":[48,52]},{"id_str":"1661523610111193088","name":"xAI","screen_name":"xAI","indices":[48,52]},{"id_str":"1661523610111193088","name":"xAI","screen_name":"xAI","indices":[48,52]}]},"favorited":false,"lang":"en","quoted_status_id_str":"1823599819551858830","quoted_status_permalink":{"url":"https://t.co/2GZhdUMNgA","expanded":"https://twitter.com/arena/status/1823599819551858830","display":"x.com/arena/status/1…"},"retweeted":false,"fact_check":null,"id":"1823615520907911400","view_count":37339,"bookmark_count":30,"created_at":1723618795000,"favorite_count":354,"quote_count":6,"reply_count":12,"retweet_count":17,"user_id_str":"952696452173594625","conversation_id_str":"1823615520907911400","full_text":"Grok-2 is here, a new frontier-level model from @xAI!\n\nI still remember the good old days when I was a GPU-poor grad student, playing with the Vicuna model and building the Chatbot Arena leaderboard with just a few GPUs.\n\nBut now, my job at xAI is developing systems for the 100K GPU cluster. If you want to be among the first to see AGI, join xAI at https://t.co/AlbqAelHiD.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[]},"display_text_range":[0,277],"lang":"en","quoted_tweet":{"id":"2001356624855023669","text":"How long have you been \"planning to understand\" how modern LLM inference works?\n\nWe just gave you a readable version of SGLang you can finish over the weekend.\n\nIntroducing mini-SGLang ⚡\n\nWe distilled SGLang from 300K into 5,000 lines. Kept the core design, cut the complexity. Without sacrificing performance — nearly identical to SGLang online.\n\nIt is built for engineers, researchers, and students who want to see how inference really works and learn better from code than papers.\n\n⭐ Star us on GitHub: https://t.co/Nk5NCXXdTz\n🧵 (1/3)","full_text":"How long have you been \"planning to understand\" how modern LLM inference works?\n\nWe just gave you a readable version of SGLang you can finish over the weekend.\n\nIntroducing mini-SGLang ⚡\n\nWe distilled SGLang from 300K into 5,000 lines. Kept the core design, cut the complexity. Without sacrificing performance — nearly identical to SGLang online.\n\nIt is built for engineers, researchers, and students who want to see how inference really works and learn better from code than papers.\n\n⭐ Star us on GitHub: https://t.co/Nk5NCXXdTz\n🧵 (1/3)","created_at":1765995577000,"author_id":"1822588444046249984","author":{"id":"1822588444046249984","name":"LMSYS Org","username":"lmsysorg","screen_name":"lmsysorg","profile_image_url":"https://pbs.twimg.com/profile_images/1823251762243121152/0cgIY6yc_400x400.jpg","profile_image_url_https":"https://pbs.twimg.com/profile_images/1823251762243121152/0cgIY6yc_400x400.jpg","is_blue_verified":1},"public_metrics":{"like_count":1158,"retweet_count":179,"reply_count":30,"quote_count":47}},"fact_check":null,"id":"2001422266769510435","view_count":34217,"bookmark_count":184,"created_at":1766011227000,"favorite_count":350,"quote_count":1,"reply_count":9,"retweet_count":29,"user_id_str":"952696452173594625","conversation_id_str":"2001422266769510435","full_text":"Industry-grade inference engines have become incredibly complex, making learning and experimentation much harder than they should be.\n\nThis project reimplements a minimal version of SGLang in ~5K lines of code. It achieves similar performance on basic benchmarks and covers the major concepts in state-of-the-art inference engines. Happy hacking!","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"rapidapi","fetched_at":1768099822792,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"bookmarked":false,"display_text_range":[0,157],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[]},"favorited":false,"lang":"en","quoted_status_id_str":"1966285780437053640","quoted_status_permalink":{"url":"https://t.co/An20eOOj9K","expanded":"https://twitter.com/lmsysorg/status/1966285780437053640","display":"x.com/lmsysorg/statu…"},"retweeted":false,"fact_check":null,"id":"1966312698100203994","view_count":28422,"bookmark_count":125,"created_at":1757640454000,"favorite_count":323,"quote_count":0,"reply_count":4,"retweet_count":27,"user_id_str":"952696452173594625","conversation_id_str":"1966312698100203994","full_text":"SGLang can now utilize CPU and external storage to reduce TTFT for your LLM queries through integration with Mooncake Storage, DeepSeek 3FS KVStore, or NIXL.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[{"id_str":"1385650483801382912","name":"Ying Sheng","screen_name":"ying11231","indices":[24,34]},{"id_str":"44196397","name":"Elon Musk","screen_name":"elonmusk","indices":[39,48]}]},"display_text_range":[0,280],"lang":"en","quoted_tweet":{"id":"1998079551369593222","text":"We've been running @radixark for a few months, started by many core developers in SGLang @lmsysorg and its extended ecosystem (slime @slime_framework , AReaL @jxwuyi). I left @xai in August — a place where I built deep emotions and countless beautiful memories. It was the best place I’ve ever worked, the place I watched grow from a few dozen people to hundreds, and it truly felt like home. What pushed me to make such a hard decision is the momentum of building SGLang open source and the mission of creating an ambitious future, within an open spirit that I learnt from my first job at @databricks after my PhD.\n\nWe started SGLang in the summer of 2023 and made it public in January 2024. Over the past 2 years, hundreds of people have made great efforts to get to where they are today. We experienced several waves of growth after its first release. I still remember the many dark nights in the summer of 2024, I spent with @lm_zheng , @lsyincs , and @zhyncs42 debugging, while @ispobaoke single-handedly took on DeepSeek inference optimizations, seeing @GenAI_is_real and the community strike team tag-teaming on-call shifts non-stop. There are so many more who have joined that I'm out of space to call out, but they're recorded on the GitHub contributor list forever. The demands grow exponentially, and we have been pushed to make it a dedicated effort supported by RadixArk. It’s the step-by-step journey of a thousand miles that has carried us here today, and the same relentless Long March that will lead us into the tens of thousands of miles yet to come.\n\nThe story never stops growing. Over the past year, we’ve seen something very clear:\nThe world is full of people eager to build AI, but the infrastructure that makes it possible is not shared. The most advanced inference and training stacks live inside a few companies. Everyone else is forced to rebuild the same schedulers, compilers, serving engines, and training pipelines again and again — often under enormous pressure, with lots of duplicated effort and wasted insight.\n\nRadixArk was born to change that. Today, we’re building an infrastructure-first, deep-tech company with a simple and ambitious mission:\n\"Make frontier-level AI infrastructure open and accessible to everyone.\"\n\nIf the two values below resonate with you, come talk to us:\n\n(1) Engineering as an art.\nInfrastructure is a first-class citizen in RadixArk. We care about elegant design and code that lasts. Beneath every line of code lies the soul of the engineer who wrote it.\n\n(2) A belief in openness.\nWe share what we build. We bet on long-term compounding through community, contribution, and giving more than we take.\n\nA product is defined by its users, yet it truly comes alive the moment functionality transcends mere utility and begins to embody aesthetics.\n\nThanks to all the miles (the name of our first released RL framework; see below).\n\nhttps://t.co/2vio4Eiiac","full_text":"We've been running @radixark for a few months, started by many core developers in SGLang @lmsysorg and its extended ecosystem (slime @slime_framework , AReaL @jxwuyi). I left @xai in August — a place where I built deep emotions and countless beautiful memories. It was the best place I’ve ever worked, the place I watched grow from a few dozen people to hundreds, and it truly felt like home. What pushed me to make such a hard decision is the momentum of building SGLang open source and the mission of creating an ambitious future, within an open spirit that I learnt from my first job at @databricks after my PhD.\n\nWe started SGLang in the summer of 2023 and made it public in January 2024. Over the past 2 years, hundreds of people have made great efforts to get to where they are today. We experienced several waves of growth after its first release. I still remember the many dark nights in the summer of 2024, I spent with @lm_zheng , @lsyincs , and @zhyncs42 debugging, while @ispobaoke single-handedly took on DeepSeek inference optimizations, seeing @GenAI_is_real and the community strike team tag-teaming on-call shifts non-stop. There are so many more who have joined that I'm out of space to call out, but they're recorded on the GitHub contributor list forever. The demands grow exponentially, and we have been pushed to make it a dedicated effort supported by RadixArk. It’s the step-by-step journey of a thousand miles that has carried us here today, and the same relentless Long March that will lead us into the tens of thousands of miles yet to come.\n\nThe story never stops growing. Over the past year, we’ve seen something very clear:\nThe world is full of people eager to build AI, but the infrastructure that makes it possible is not shared. The most advanced inference and training stacks live inside a few companies. Everyone else is forced to rebuild the same schedulers, compilers, serving engines, and training pipelines again and again — often under enormous pressure, with lots of duplicated effort and wasted insight.\n\nRadixArk was born to change that. Today, we’re building an infrastructure-first, deep-tech company with a simple and ambitious mission:\n\"Make frontier-level AI infrastructure open and accessible to everyone.\"\n\nIf the two values below resonate with you, come talk to us:\n\n(1) Engineering as an art.\nInfrastructure is a first-class citizen in RadixArk. We care about elegant design and code that lasts. Beneath every line of code lies the soul of the engineer who wrote it.\n\n(2) A belief in openness.\nWe share what we build. We bet on long-term compounding through community, contribution, and giving more than we take.\n\nA product is defined by its users, yet it truly comes alive the moment functionality transcends mere utility and begins to embody aesthetics.\n\nThanks to all the miles (the name of our first released RL framework; see below).\n\nhttps://t.co/2vio4Eiiac","created_at":1765214262000,"author_id":"1385650483801382912","author":{"id":"1385650483801382912","name":"Ying Sheng","username":"ying11231","screen_name":"ying11231","profile_image_url":"https://pbs.twimg.com/profile_images/1546324576681873408/fIor2sAZ_400x400.jpg","profile_image_url_https":"https://pbs.twimg.com/profile_images/1546324576681873408/fIor2sAZ_400x400.jpg","is_blue_verified":1},"public_metrics":{"like_count":1090,"retweet_count":123,"reply_count":110,"quote_count":45}},"fact_check":null,"id":"1998745512875602347","view_count":29745,"bookmark_count":46,"created_at":1765373039000,"favorite_count":296,"quote_count":0,"reply_count":4,"retweet_count":8,"user_id_str":"952696452173594625","conversation_id_str":"1998745512875602347","full_text":"I’ve learned a ton from @ying11231 and @elonmusk this year. The way they think and the instincts they rely on when tackling challenges are insanely similar.\n\nOnly when you work closely with them do you start to realize just how visionary, consistent, and truth-seeking they really are.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"rapidapi","fetched_at":1767495081514,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"bookmarked":false,"display_text_range":[0,256],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[{"id_str":"1891523745929732096","name":"Thinking Machines","screen_name":"thinkymachines","indices":[128,143]}]},"favorited":false,"lang":"en","quoted_status_id_str":"1970240927429206161","quoted_status_permalink":{"url":"https://t.co/VYBUV75SRb","expanded":"https://twitter.com/lmsysorg/status/1970240927429206161","display":"x.com/lmsysorg/statu…"},"retweeted":false,"fact_check":null,"id":"1970256400329953371","view_count":26648,"bookmark_count":84,"created_at":1758580706000,"favorite_count":288,"quote_count":1,"reply_count":5,"retweet_count":30,"user_id_str":"952696452173594625","conversation_id_str":"1970256400329953371","full_text":"The main branch of sglang now supports deterministic inference with user-specified per-request seeds!\n\nIt utilized kernels from @thinkymachines and introduced new optimizations & coverage. Run out of box for most hardware backends and pytorch versions.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[]},"display_text_range":[0,101],"lang":"en","quoted_tweet":{"id":"1995553798274912390","text":"Releasing Alpha-MoE: Megakernel for fast Tensor Parallel Inference!\n\nUp to 200% faster execution of MoE layer in SGLang, with 17% higher average throughput on Qwen3-Next-80B, and 10% higher average throughput on DeepSeek \n\nProud to showcase my recent work at @Aleph__Alpha🧵 https://t.co/QLDOwhF4oC","full_text":"Releasing Alpha-MoE: Megakernel for fast Tensor Parallel Inference!\n\nUp to 200% faster execution of MoE layer in SGLang, with 17% higher average throughput on Qwen3-Next-80B, and 10% higher average throughput on DeepSeek \n\nProud to showcase my recent work at @Aleph__Alpha🧵 https://t.co/QLDOwhF4oC","created_at":1764612075000,"author_id":"1911051072468312064","author":{"id":"1911051072468312064","name":"SzymonOzog (NeurIPS)","username":"SzymonOzog_","screen_name":"SzymonOzog_","profile_image_url":"https://pbs.twimg.com/profile_images/1914268416115912705/TseyFx-Z_400x400.jpg","profile_image_url_https":"https://pbs.twimg.com/profile_images/1914268416115912705/TseyFx-Z_400x400.jpg","is_blue_verified":null},"public_metrics":{"like_count":315,"retweet_count":33,"reply_count":12,"quote_count":9}},"fact_check":null,"id":"1995677691513864555","view_count":26404,"bookmark_count":130,"created_at":1764641614000,"favorite_count":269,"quote_count":0,"reply_count":4,"retweet_count":13,"user_id_str":"952696452173594625","conversation_id_str":"1995677691513864555","full_text":"Very interesting optimizations to push the limit of GPUs. The team will test and try to integrate it.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":1766049241625,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"bookmarked":false,"display_text_range":[0,38],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[]},"favorited":false,"lang":"en","quoted_status_id_str":"1990530499752980638","quoted_status_permalink":{"url":"https://t.co/tBMdxOXOdh","expanded":"https://twitter.com/xai/status/1990530499752980638","display":"x.com/xai/status/199…"},"retweeted":false,"fact_check":null,"id":"1990531757108810208","view_count":29937,"bookmark_count":9,"created_at":1763414727000,"favorite_count":252,"quote_count":0,"reply_count":12,"retweet_count":10,"user_id_str":"952696452173594625","conversation_id_str":"1990531757108810208","full_text":"Grok 4.1 is out! Amazing improvements.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"bookmarked":false,"display_text_range":[0,214],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[]},"favorited":false,"lang":"en","quoted_status_id_str":"1981103763441250387","quoted_status_permalink":{"url":"https://t.co/XukMNG3ebz","expanded":"https://twitter.com/lmsysorg/status/1981103763441250387","display":"x.com/lmsysorg/statu…"},"retweeted":false,"fact_check":null,"id":"1981106698258694510","view_count":24438,"bookmark_count":112,"created_at":1761167618000,"favorite_count":247,"quote_count":0,"reply_count":4,"retweet_count":27,"user_id_str":"952696452173594625","conversation_id_str":"1981106698258694510","full_text":"MoE exposed interesting opportunities to fully utilize the heterogeneous hardware resources (CPU + GPU).\n\nKTransformer team is upstreaming their cool optimizations into the sglang stack to combine the best of both.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"bookmarked":false,"display_text_range":[0,142],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[]},"favorited":false,"lang":"en","quoted_status_id_str":"1827047684714463603","quoted_status_permalink":{"url":"https://t.co/wGFFQ8uHnV","expanded":"https://twitter.com/ibab/status/1827047684714463603","display":"x.com/ibab/status/18…"},"retweeted":false,"fact_check":null,"id":"1827369068380193091","view_count":27141,"bookmark_count":13,"created_at":1724513711000,"favorite_count":226,"quote_count":2,"reply_count":14,"retweet_count":21,"user_id_str":"952696452173594625","conversation_id_str":"1827369068380193091","full_text":"This is the xAI speed over the past few weeks: 1 Elo point higher per day in model quality, and 10 tokens/s faster per day in inference speed.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0}],"ctweets":[{"bookmarked":false,"display_text_range":[0,31],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[{"display_url":"x.com/i/events/19427…","expanded_url":"https://x.com/i/events/1942716886258528256","url":"https://t.co/Ody4Uwh1n2","indices":[8,31]}],"user_mentions":[]},"favorited":false,"lang":"en","possibly_sensitive":false,"possibly_sensitive_editable":true,"retweeted":false,"fact_check":null,"id":"1943153321801633805","view_count":84502,"bookmark_count":39,"created_at":1752118828000,"favorite_count":501,"quote_count":5,"reply_count":68,"retweet_count":38,"user_id_str":"952696452173594625","conversation_id_str":"1943153321801633805","full_text":"Grok4 🚀\nhttps://t.co/Ody4Uwh1n2","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":0,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[{"id_str":"44196397","name":"Elon Musk","screen_name":"elonmusk","indices":[56,65]}]},"display_text_range":[0,114],"lang":"en","quoted_tweet":{"id":"2008683630768648493","text":"This is my favorite clip of the new Elon pod. He opens up saying xAI struggles with memory usage/bandwidth and CUDA kernel optimization (matmul, attention, MoE, etc). If you are good kernel or performance engineering in general, you should apply. Steer the world in a better direction.","full_text":"This is my favorite clip of the new Elon pod. He opens up saying xAI struggles with memory usage/bandwidth and CUDA kernel optimization (matmul, attention, MoE, etc). If you are good kernel or performance engineering in general, you should apply. Steer the world in a better direction.","created_at":1767742471000,"author_id":"1595935520999510016","author":{"id":"1595935520999510016","name":"Elliot Arledge","username":"elliotarledge","screen_name":"elliotarledge","profile_image_url":"https://pbs.twimg.com/profile_images/2005693506644574208/Hp8sqTPA_400x400.jpg","profile_image_url_https":"https://pbs.twimg.com/profile_images/2005693506644574208/Hp8sqTPA_400x400.jpg","is_blue_verified":1},"public_metrics":{"like_count":906,"retweet_count":30,"reply_count":34,"quote_count":14}},"fact_check":null,"id":"2009196655288217735","view_count":90962,"bookmark_count":104,"created_at":1767864786000,"favorite_count":538,"quote_count":4,"reply_count":35,"retweet_count":39,"user_id_str":"952696452173594625","conversation_id_str":"2009196655288217735","full_text":"Apply to the kernel team and you get the chance to show @elonmusk how to write efficient kernels in PTX for GB300.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"rapidapi","fetched_at":1770137356512,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"bookmarked":false,"display_text_range":[0,155],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[{"id_str":"1661523610111193088","name":"xAI","screen_name":"xai","indices":[135,139]}]},"favorited":false,"lang":"en","quoted_status_id_str":"1891706264800936307","quoted_status_permalink":{"url":"https://t.co/fhF5Mm3HQg","expanded":"https://twitter.com/arena/status/1891706264800936307","display":"x.com/arena/status/1…"},"retweeted":false,"fact_check":null,"id":"1891713559874187417","view_count":28678,"bookmark_count":25,"created_at":1739854633000,"favorite_count":699,"quote_count":6,"reply_count":21,"retweet_count":51,"user_id_str":"952696452173594625","conversation_id_str":"1891713559874187417","full_text":"People have been wondering which model will be the first to reach a 1400 Elo score; no one believed it would be Grok one year ago, but @xAI made it happen!","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[]},"display_text_range":[0,262],"lang":"en","quoted_tweet":{"id":"1990866667237191758","text":"Never block the GPU!\n\nIn a new @modal blogpost, we walk through a major class of inefficiency in AI inference: host overhead.\n\nWe include three cases where we worked with @sgl_project to cut host overhead and prevent GPU stalls.\n\nEvery microsecond counts.\n\nhttps://t.co/ZeumrZpSKE","full_text":"Never block the GPU!\n\nIn a new @modal blogpost, we walk through a major class of inefficiency in AI inference: host overhead.\n\nWe include three cases where we worked with @sgl_project to cut host overhead and prevent GPU stalls.\n\nEvery microsecond counts.\n\nhttps://t.co/ZeumrZpSKE","created_at":1763494576000,"author_id":"950189417754513409","author":{"id":"950189417754513409","name":"Charles 🎉 Frye","username":"charles_irl","screen_name":"charles_irl","profile_image_url":"https://pbs.twimg.com/profile_images/1320840184494108674/d7A64nIG_400x400.jpg","profile_image_url_https":"https://pbs.twimg.com/profile_images/1320840184494108674/d7A64nIG_400x400.jpg","is_blue_verified":1},"public_metrics":{"like_count":416,"retweet_count":40,"reply_count":6,"quote_count":10}},"fact_check":null,"id":"1992497243686080769","view_count":103176,"bookmark_count":277,"created_at":1763883336000,"favorite_count":676,"quote_count":2,"reply_count":21,"retweet_count":30,"user_id_str":"952696452173594625","conversation_id_str":"1992497243686080769","full_text":"Fun fact:\nWhile inference is running on the GPU, I probably spent more time optimizing CPU overhead than the GPU kernels themselves during my development of SGLang.\n\nMany big refactors and small details were needed to achieve a zero-CPU-overhead batch scheduler.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":1765371346955,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"bookmarked":false,"display_text_range":[0,200],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[]},"favorited":false,"lang":"en","quoted_status_id_str":"1872242657348710721","quoted_status_permalink":{"url":"https://t.co/jq2TrcpMiA","expanded":"https://twitter.com/deepseek_ai/status/1872242657348710721","display":"x.com/deepseek_ai/st…"},"retweeted":false,"fact_check":null,"id":"1872266844603904398","view_count":43801,"bookmark_count":55,"created_at":1735218175000,"favorite_count":515,"quote_count":5,"reply_count":18,"retweet_count":37,"user_id_str":"952696452173594625","conversation_id_str":"1872266844603904398","full_text":"Highly respected! It is so impressive given the results and the very limited resources they have compared to other big labs.\n\n\"DeepSeek-V3 is trained on a cluster equipped with 2048 NVIDIA H800 GPUs.\"","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"bookmarked":false,"display_text_range":[0,142],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[]},"favorited":false,"lang":"en","quoted_status_id_str":"1827047684714463603","quoted_status_permalink":{"url":"https://t.co/wGFFQ8uHnV","expanded":"https://twitter.com/ibab/status/1827047684714463603","display":"x.com/ibab/status/18…"},"retweeted":false,"fact_check":null,"id":"1827369068380193091","view_count":27141,"bookmark_count":13,"created_at":1724513711000,"favorite_count":226,"quote_count":2,"reply_count":14,"retweet_count":21,"user_id_str":"952696452173594625","conversation_id_str":"1827369068380193091","full_text":"This is the xAI speed over the past few weeks: 1 Elo point higher per day in model quality, and 10 tokens/s faster per day in inference speed.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"bookmarked":false,"display_text_range":[0,279],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[{"display_url":"x.ai/careers","expanded_url":"https://x.ai/careers","url":"https://t.co/AlbqAelHiD","indices":[351,374]},{"display_url":"x.ai/careers","expanded_url":"https://x.ai/careers","url":"https://t.co/AlbqAelHiD","indices":[351,374]}],"user_mentions":[{"id_str":"1661523610111193088","name":"xAI","screen_name":"xai","indices":[48,52]},{"id_str":"1661523610111193088","name":"xAI","screen_name":"xAI","indices":[48,52]},{"id_str":"1661523610111193088","name":"xAI","screen_name":"xAI","indices":[48,52]}]},"favorited":false,"lang":"en","quoted_status_id_str":"1823599819551858830","quoted_status_permalink":{"url":"https://t.co/2GZhdUMNgA","expanded":"https://twitter.com/arena/status/1823599819551858830","display":"x.com/arena/status/1…"},"retweeted":false,"fact_check":null,"id":"1823615520907911400","view_count":37339,"bookmark_count":30,"created_at":1723618795000,"favorite_count":354,"quote_count":6,"reply_count":12,"retweet_count":17,"user_id_str":"952696452173594625","conversation_id_str":"1823615520907911400","full_text":"Grok-2 is here, a new frontier-level model from @xAI!\n\nI still remember the good old days when I was a GPU-poor grad student, playing with the Vicuna model and building the Chatbot Arena leaderboard with just a few GPUs.\n\nBut now, my job at xAI is developing systems for the 100K GPU cluster. If you want to be among the first to see AGI, join xAI at https://t.co/AlbqAelHiD.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"bookmarked":false,"display_text_range":[0,38],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[]},"favorited":false,"lang":"en","quoted_status_id_str":"1990530499752980638","quoted_status_permalink":{"url":"https://t.co/tBMdxOXOdh","expanded":"https://twitter.com/xai/status/1990530499752980638","display":"x.com/xai/status/199…"},"retweeted":false,"fact_check":null,"id":"1990531757108810208","view_count":29937,"bookmark_count":9,"created_at":1763414727000,"favorite_count":252,"quote_count":0,"reply_count":12,"retweet_count":10,"user_id_str":"952696452173594625","conversation_id_str":"1990531757108810208","full_text":"Grok 4.1 is out! Amazing improvements.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"bookmarked":false,"display_text_range":[0,280],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[]},"favorited":false,"lang":"en","quoted_status_id_str":"1866224902711411050","quoted_status_permalink":{"url":"https://t.co/B16xxZBptv","expanded":"https://twitter.com/xai/status/1866224902711411050","display":"x.com/xai/status/186…"},"retweeted":false,"fact_check":null,"id":"1866280165254263115","view_count":132795,"bookmark_count":79,"created_at":1733790839000,"favorite_count":529,"quote_count":4,"reply_count":10,"retweet_count":27,"user_id_str":"952696452173594625","conversation_id_str":"1866280165254263115","full_text":"We released a new image generation model that was trained from scratch at xAI!\n\nIt has been fun optimizing the inference for it. Because it's an autoregressive model, we can easily port it to SGLang and reuse all the great optimizations like the new zero-overhead batch scheduler, torch.compile, and many custom kernels.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[]},"display_text_range":[0,277],"lang":"en","quoted_tweet":{"id":"2001356624855023669","text":"How long have you been \"planning to understand\" how modern LLM inference works?\n\nWe just gave you a readable version of SGLang you can finish over the weekend.\n\nIntroducing mini-SGLang ⚡\n\nWe distilled SGLang from 300K into 5,000 lines. Kept the core design, cut the complexity. Without sacrificing performance — nearly identical to SGLang online.\n\nIt is built for engineers, researchers, and students who want to see how inference really works and learn better from code than papers.\n\n⭐ Star us on GitHub: https://t.co/Nk5NCXXdTz\n🧵 (1/3)","full_text":"How long have you been \"planning to understand\" how modern LLM inference works?\n\nWe just gave you a readable version of SGLang you can finish over the weekend.\n\nIntroducing mini-SGLang ⚡\n\nWe distilled SGLang from 300K into 5,000 lines. Kept the core design, cut the complexity. Without sacrificing performance — nearly identical to SGLang online.\n\nIt is built for engineers, researchers, and students who want to see how inference really works and learn better from code than papers.\n\n⭐ Star us on GitHub: https://t.co/Nk5NCXXdTz\n🧵 (1/3)","created_at":1765995577000,"author_id":"1822588444046249984","author":{"id":"1822588444046249984","name":"LMSYS Org","username":"lmsysorg","screen_name":"lmsysorg","profile_image_url":"https://pbs.twimg.com/profile_images/1823251762243121152/0cgIY6yc_400x400.jpg","profile_image_url_https":"https://pbs.twimg.com/profile_images/1823251762243121152/0cgIY6yc_400x400.jpg","is_blue_verified":1},"public_metrics":{"like_count":1158,"retweet_count":179,"reply_count":30,"quote_count":47}},"fact_check":null,"id":"2001422266769510435","view_count":34217,"bookmark_count":184,"created_at":1766011227000,"favorite_count":350,"quote_count":1,"reply_count":9,"retweet_count":29,"user_id_str":"952696452173594625","conversation_id_str":"2001422266769510435","full_text":"Industry-grade inference engines have become incredibly complex, making learning and experimentation much harder than they should be.\n\nThis project reimplements a minimal version of SGLang in ~5K lines of code. It achieves similar performance on basic benchmarks and covers the major concepts in state-of-the-art inference engines. Happy hacking!","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"rapidapi","fetched_at":1768099822792,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[]},"display_text_range":[0,280],"lang":"en","quoted_tweet":{"id":"2011869695482249680","text":"Introducing SGLang Pipeline Parallelism: The highest-performance open-source PP implementation for 1M+ token contexts.\n\nWe’ve engineered a robust multi-node scaling solution that redefines efficiency: \n✨ Dynamic Chunking: Intelligently minimizes pipeline bubbles. \n⚡ Async P2P Communication: Maximize hardware utilization. \n🧩 Seamless Compatibility: Works perfectly with PD Disaggregation, HiCache, and hybrid parallelism.\n\nAchieve 3.31x throughput gains (vs TP8) and unlock the full potential of your trillion-parameter models. 👇\n\n#SGLang #LLM #DeepSeek #AIInference #OpenSource","full_text":"Introducing SGLang Pipeline Parallelism: The highest-performance open-source PP implementation for 1M+ token contexts.\n\nWe’ve engineered a robust multi-node scaling solution that redefines efficiency: \n✨ Dynamic Chunking: Intelligently minimizes pipeline bubbles. \n⚡ Async P2P Communication: Maximize hardware utilization. \n🧩 Seamless Compatibility: Works perfectly with PD Disaggregation, HiCache, and hybrid parallelism.\n\nAchieve 3.31x throughput gains (vs TP8) and unlock the full potential of your trillion-parameter models. 👇\n\n#SGLang #LLM #DeepSeek #AIInference #OpenSource","created_at":1768502088000,"author_id":"1822588444046249984","author":{"id":"1822588444046249984","name":"LMSYS Org","username":"lmsysorg","screen_name":"lmsysorg","profile_image_url":"https://pbs.twimg.com/profile_images/1823251762243121152/0cgIY6yc_400x400.jpg","profile_image_url_https":"https://pbs.twimg.com/profile_images/1823251762243121152/0cgIY6yc_400x400.jpg","is_blue_verified":1},"public_metrics":{"like_count":176,"retweet_count":23,"reply_count":3,"quote_count":8}},"fact_check":null,"id":"2012069688122003890","view_count":35150,"bookmark_count":206,"created_at":1768549770000,"favorite_count":364,"quote_count":2,"reply_count":9,"retweet_count":34,"user_id_str":"952696452173594625","conversation_id_str":"2012069688122003890","full_text":"Chunked pipeline parallelism is arguably the most general and scalable system technique for accelerating super-long-context inference.\n\nIt remains underrated today, largely because there still isn’t a strong, high-quality open-source implementation.\nThe SGLang team recently fully optimized it and published a detailed blog post explaining all the key details.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"rapidapi","fetched_at":1770776994668,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"bookmarked":false,"display_text_range":[0,266],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[{"id_str":"61559439","name":"NVIDIA","screen_name":"nvidia","indices":[99,106]},{"id_str":"1603444903","name":"yv","screen_name":"yvbbrjdr","indices":[240,249]},{"id_str":"1524622862040018944","name":"Richard Chen","screen_name":"richardczl","indices":[254,265]}]},"favorited":false,"lang":"en","quoted_status_id_str":"1977891284460384465","quoted_status_permalink":{"url":"https://t.co/T3yh6uNacp","expanded":"https://twitter.com/lmsysorg/status/1977891284460384465","display":"x.com/lmsysorg/statu…"},"retweeted":false,"fact_check":null,"id":"1977893977895395390","view_count":12675,"bookmark_count":23,"created_at":1760401646000,"favorite_count":173,"quote_count":0,"reply_count":8,"retweet_count":12,"user_id_str":"952696452173594625","conversation_id_str":"1977893977895395390","full_text":"Personal AI computing has just become more accessible than ever before with the new DGX Spark from @nvidia - a tiny yet powerful Blackwell GPU in your hand.\n\nCheck out the firsthand unboxing video and blog from the SGLang community members @yvbbrjdr and @richardczl.","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"bookmarked":false,"display_text_range":[0,140],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[{"display_url":"x.com/i/grok/share/U…","expanded_url":"https://x.com/i/grok/share/Ukl19CIFqyHbdKBFETw8Mevjx","url":"https://t.co/bQcKAl2yQf","indices":[117,140]}],"user_mentions":[]},"favorited":false,"lang":"en","possibly_sensitive":false,"possibly_sensitive_editable":true,"retweeted":false,"fact_check":null,"id":"1863527658803663095","view_count":7670,"bookmark_count":4,"created_at":1733134591000,"favorite_count":68,"quote_count":0,"reply_count":7,"retweet_count":1,"user_id_str":"952696452173594625","conversation_id_str":"1863527658803663095","full_text":"The Grok personalization looks cool! It might not work on the first try, but it finally worked after a few prompts.\n\nhttps://t.co/bQcKAl2yQf","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":0,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"bookmarked":false,"display_text_range":[0,66],"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[]},"favorited":false,"lang":"en","quoted_status_id_str":"1971351945391824999","quoted_status_permalink":{"url":"https://t.co/knAyOHne3K","expanded":"https://twitter.com/lmsysorg/status/1971351945391824999","display":"x.com/lmsysorg/statu…"},"retweeted":false,"fact_check":null,"id":"1971397642073846248","view_count":15963,"bookmark_count":35,"created_at":1758852799000,"favorite_count":197,"quote_count":1,"reply_count":6,"retweet_count":10,"user_id_str":"952696452173594625","conversation_id_str":"1971397642073846248","full_text":"Rack-scale inference is the future, and the team keeps pushing it!","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0},{"entities":{"hashtags":[],"symbols":[],"timestamps":[],"urls":[],"user_mentions":[{"id_str":"25401953","indices":[203,212],"name":"Peter Steinberger 🦞","screen_name":"steipete"}]},"display_text_range":[0,275],"lang":"en","quoted_tweet":{"id":"2032614332572533222","text":"🎉 SGLang is now a supported model provider in @OpenClaw!\nSGLang serves trillions of tokens/day across 400K+ GPUs. Now your local deployment is first-class in OpenClaw too.🦞 https://t.co/DmZyoBiXVa","full_text":"🎉 SGLang is now a supported model provider in @OpenClaw!\nSGLang serves trillions of tokens/day across 400K+ GPUs. Now your local deployment is first-class in OpenClaw too.🦞 https://t.co/DmZyoBiXVa","created_at":1773447995000,"author_id":"1822588444046249984","author":{"id":"1822588444046249984","name":"LMSYS Org","username":"lmsysorg","screen_name":"lmsysorg","profile_image_url":"https://pbs.twimg.com/profile_images/1823251762243121152/0cgIY6yc_400x400.jpg","profile_image_url_https":"https://pbs.twimg.com/profile_images/1823251762243121152/0cgIY6yc_400x400.jpg","is_blue_verified":1},"public_metrics":{"like_count":71,"retweet_count":18,"reply_count":5,"quote_count":5}},"fact_check":null,"id":"2032665982356041747","view_count":49915,"bookmark_count":43,"created_at":1773460309000,"favorite_count":150,"quote_count":0,"reply_count":6,"retweet_count":10,"user_id_str":"952696452173594625","conversation_id_str":"2032665982356041747","full_text":"OpenClaw now uses a refactored plugin system to support SGLang as a native provider.\n\nInitially, I submitted a PR to add support. It had minimal changes, but I wasn't entirely happy with the structure.\n\n@steipete quickly followed up with a refactor that perfectly matches the design I had in mind. Respect! 🫡","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":1,"is_ai":null,"ai_score":null,"source":"koosocial_engage","fetched_at":1774564542787,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":1773720017396,"poll_count":1,"poll_complete":1},{"bookmarked":false,"display_text_range":[0,276],"entities":{"hashtags":[{"indices":[3,15],"text":"NeurIPS2024"},{"indices":[3,15],"text":"NeurIPS2024"}],"media":[{"display_url":"pic.x.com/FhiGWyq9qB","expanded_url":"https://x.com/lm_zheng/status/1866574315904245767/photo/1","id_str":"1866573848558055424","indices":[277,300],"media_key":"3_1866573848558055424","media_url_https":"https://pbs.twimg.com/media/GednBpZWMAAEuqJ.jpg","type":"photo","url":"https://t.co/FhiGWyq9qB","ext_media_availability":{"status":"Available"},"features":{"large":{"faces":[]},"medium":{"faces":[]},"small":{"faces":[]},"orig":{"faces":[]}},"sizes":{"large":{"h":946,"w":952,"resize":"fit"},"medium":{"h":946,"w":952,"resize":"fit"},"small":{"h":676,"w":680,"resize":"fit"},"thumb":{"h":150,"w":150,"resize":"crop"}},"original_info":{"height":946,"width":952,"focus_rects":[{"x":0,"y":413,"w":952,"h":533},{"x":3,"y":0,"w":946,"h":946},{"x":61,"y":0,"w":830,"h":946},{"x":240,"y":0,"w":473,"h":946},{"x":0,"y":0,"w":952,"h":946}]},"allow_download_status":{"allow_download":true},"media_results":{"result":{"media_key":"3_1866573848558055424"}}}],"symbols":[],"timestamps":[],"urls":[{"display_url":"neurips.cc/virtual/2024/p…","expanded_url":"https://neurips.cc/virtual/2024/poster/94872","url":"https://t.co/a1g6iMdsja","indices":[277,300]}],"user_mentions":[]},"extended_entities":{"media":[{"display_url":"pic.x.com/FhiGWyq9qB","expanded_url":"https://x.com/lm_zheng/status/1866574315904245767/photo/1","id_str":"1866573848558055424","indices":[277,300],"media_key":"3_1866573848558055424","media_url_https":"https://pbs.twimg.com/media/GednBpZWMAAEuqJ.jpg","type":"photo","url":"https://t.co/FhiGWyq9qB","ext_media_availability":{"status":"Available"},"features":{"large":{"faces":[]},"medium":{"faces":[]},"small":{"faces":[]},"orig":{"faces":[]}},"sizes":{"large":{"h":946,"w":952,"resize":"fit"},"medium":{"h":946,"w":952,"resize":"fit"},"small":{"h":676,"w":680,"resize":"fit"},"thumb":{"h":150,"w":150,"resize":"crop"}},"original_info":{"height":946,"width":952,"focus_rects":[{"x":0,"y":413,"w":952,"h":533},{"x":3,"y":0,"w":946,"h":946},{"x":61,"y":0,"w":830,"h":946},{"x":240,"y":0,"w":473,"h":946},{"x":0,"y":0,"w":952,"h":946}]},"allow_download_status":{"allow_download":true},"media_results":{"result":{"media_key":"3_1866573848558055424"}}}]},"favorited":false,"lang":"en","possibly_sensitive":false,"possibly_sensitive_editable":true,"retweeted":false,"fact_check":null,"id":"1866574315904245767","view_count":15822,"bookmark_count":47,"created_at":1733860970000,"favorite_count":218,"quote_count":1,"reply_count":5,"retweet_count":22,"user_id_str":"952696452173594625","conversation_id_str":"1866574315904245767","full_text":"At #NeurIPS2024 this week. Happy to chat about anything related to LLMs and xAI: our 100K training cluster, inference stack, Grok3, and the newly released AR image gen model.\n\nWe're presenting an SGLang poster on Thursday afternoon. Welcome to stop by and join the discussion:\nhttps://t.co/a1g6iMdsja","in_reply_to_user_id_str":null,"in_reply_to_status_id_str":null,"is_quote_status":0,"is_ai":null,"ai_score":null,"source":"scraping","fetched_at":null,"edit_history_tweet_ids":null,"poll_10min_at":null,"poll_3day_at":null,"poll_count":0,"poll_complete":0}],"activities":null,"interactions":null,"interactions_updated":null,"created":1774564657559,"updated":1774564657559,"type":"the innovator","hits":1},"people":[{"user":{"id":"361860294","name":"ria bhutoria","description":"stablecoins @stripe","followers_count":15790,"friends_count":2592,"statuses_count":3397,"profile_image_url_https":"https://pbs.twimg.com/profile_images/1963391844441886723/Uqx0rsV3_normal.jpg","screen_name":"riabhutoria","location":"","entities":{"description":{"urls":[]}}},"details":{"type":"The Innovator","description":"Ria Bhutoria is a product-minded payments and crypto operator who’s obsessed with making stablecoins practical, safe, and widely useful. She builds infrastructure, writes smart threads, and moves between startups and platform-level roles with infectious enthusiasm. Expect clear takes, design-first thinking, and a knack for turning research into real products.","purpose":"To accelerate mainstream adoption of stablecoins and crypto-native payments by building rigorous, user-first infrastructure and tools that make money movement faster, cheaper, and more accessible for everyone.","beliefs":"Product craft and engineering rigor beat hype; interoperability, neutrality, and sound guardrails are essential for crypto to serve real users; verification and careful design are the antidotes to blind trust; permissionless options and pragmatic rails can coexist.","facts":"Fun fact: Ria went all in on crypto seven years ago and has since parlayed that conviction into leadership bets—joining teams like Stripe and delta to ship stablecoin-first infrastructure (and she signed the Stripe announcement with a cowboy hat emoji 🤠).","strength":"Combines deep technical/product intuition with an operator’s execution muscle; great at explaining complex protocol tradeoffs in plain language; credible network and the ability to rally engineering and biz teams around infrastructure projects.","weakness":"Can sound delightfully nerdy to outsiders—sometimes assuming audiences share a high baseline of technical knowledge—and can be impatient with slow consensus or purely theoretical debates that don’t lead to product outcomes.","roast":"Ria will design the perfect, elegantly peer-reviewed stablecoin pipeline, document it in three threads, and quietly join a startup that makes it slightly less theoretical—because she can’t resist building the thing she just spent a week arguing about on Twitter.","win":"Turning early crypto conviction into measurable impact: leadership roles at high-profile projects (Stripe’s stablecoin efforts and delta as COO) and visible community influence that drives hiring, product adoption, and funding interest.","recommendation":"On X, lean into explainer threads that break down one payment or stablecoin flow per thread, publish short post-mortems of design choices, run monthly AMAs or Spaces with payments folks and merchants, use simple visuals to illustrate flows, and tag/engage people who can validate economics—this turns technical credibility into broader, practical influence."},"created":1774570442661,"type":"the innovator","id":"riabhutoria"},{"user":{"id":"1001797450825388032","name":"Oğuz Yağız Kara","description":"Co-Founder & Designer @LueStudio","followers_count":45249,"friends_count":1375,"statuses_count":6693,"profile_image_url_https":"https://pbs.twimg.com/profile_images/1958503624784400384/W2P6bsFH_normal.jpg","screen_name":"oguzyagizkara","location":"","entities":{"description":{"urls":[]},"url":{"urls":[{"display_url":"lue.studio","expanded_url":"https://lue.studio","indices":[0,23],"url":"https://t.co/ccBgUB6DHs"}]}}},"details":{"type":"The Innovator","description":"Oğuz Yağız Kara is a designer-founder who prototypes tomorrow’s products today, blending polished UI craft with bold AI-first ideas. As Co-Founder & Designer at LueStudio and a new member of Shopify’s design team, he mixes product launches, hot takes, and personal wins into a magnetic Twitter presence. His feed is equal parts demo reel, thought experiment, and life update.","purpose":"To make advanced technology feel human and usable by turning speculative ideas into tangible prototypes and products that people can actually play with. He aims to accelerate design-led innovation and inspire other makers to experiment loudly and iterate quickly.","beliefs":"Values rapid experimentation, clarity of craft, and shipping over perfection. Believes powerful tech (especially AI) should be accessible and integrated thoughtfully into design, and that honest public iteration builds better products and communities.","facts":"Fun fact: One of his tweets about a \"universal key for instant AI access\" hit over 5.6 million views. He also announced two major life moments on Twitter—his marriage and joining Shopify—both of which got huge engagement.","strength":"Visionary idea generation, fast prototyping (Figma wizardry), strong personal brand with high audience engagement, and the credibility of founder + product experience that turns attention into opportunity.","weakness":"Can be impulsive with hot takes that spark big reply storms, risks spreading attention across too many experiments, and sometimes values novelty over finishing long, deep projects.","recommendation":"Pin a short pinned-thread walkthrough for your biggest projects (screenshots, quick gifs, “what I learned” bullets). Post short prototype clips and Figma embeds, host periodic Spaces or AMAs to convert followers into loyal fans, collaborate with other designers and builders, and reuse viral moments into a newsletter or case study to deepen engagement.","roast":"You prototype faster than most people change their profile pics — your Figma files have more side projects than commitments, but hey, at least you managed to marry someone and join Shopify before your next dropdown animation was finished.","win":"Landing a role on Shopify’s design team while building a personal brand that produced a mega-viral AI tweet (5.6M+ views) — proof your ideas attract both hearts and offers."},"created":1774569827924,"type":"the innovator","id":"oguzyagizkara"},{"user":{"id":"1921674544999489536","name":"Nate Herk","description":"Founder & CEO @ Uppit AI","followers_count":3894,"friends_count":13,"statuses_count":42,"profile_image_url_https":"https://pbs.twimg.com/profile_images/1993374020436201472/e7PAqmOm_normal.jpg","screen_name":"nateherk","location":"","entities":{"description":{"urls":[]},"url":{"urls":[{"display_url":"youtube.com/@nateherk","expanded_url":"https://www.youtube.com/@nateherk","indices":[0,23],"url":"https://t.co/V3Op8W9oKF"}]}}},"details":{"type":"The Innovator","description":"Nate Herk is the Founder & CEO of Uppit AI who builds and ships practical AI agent systems that automate media, content, and lead workflows. He demonstrates fast, hands-on builds (often in under an hour) and shares free resources so others can copy his work. Nate turns complex automation into plug-and-play tools for makers and businesses.","purpose":"To democratize powerful AI automation by building accessible, repeatable agent systems that let creators and founders ship faster, scale media production, and convert attention into real business outcomes.","beliefs":"Open sharing accelerates innovation; practical demos beat theory; automation should give people control (traceability & subagents matter); growth comes from shipping useful tools and teaching others how to use them.","facts":"Fun fact: Nate famously built a multi-agent media system in about an hour and gives the system away for free. Also, despite being a startup CEO, he only follows 13 people and has tweeted just 42 times — quality over quantity.","strength":"Rapid prototyping and shipping, deep technical skill with agent architectures and integrations (n8n, Base44), clear how-to content and free resources that drive adoption, and credibility as a founder who builds working systems.","weakness":"Can under-invest in personal storytelling and audience engagement (low follow count and modest tweet volume); sometimes the product-first approach misses broader narrative hooks that grow a bigger audience faster.","roast":"You build entire armies of AI agents in an hour but follow 13 people — congrats, you're either a mastermind or running the world's smallest, most exclusive fan club.","win":"Built and published a complete suite of AI media agents (with free resources) that demo real workflows and drove major milestone celebrations (posts thanking communities for 200k and 400k), proving product traction and community interest.","recommendation":"Post consistent, high-signal threads that break down one build step-by-step; pin a clear demo + free download; publish short demo videos and X-native clips showing before/after metrics; host AMAs or Spaces to convert curious viewers into followers; reply to comments with mini-tutorials to boost engagement; strategically follow and collaborate with 10–20 complementary creators to unlock network effects."},"created":1774565135347,"type":"the innovator","id":"nateherk"},{"user":{"id":"7205472","name":"Zhixiong Pan","description":"A real cyborg vibe coder | Study @ChainFeedsxyz | ex Research @ChainNewscom, @MyToken, @IBM | Holding and accumulating BTC since 2011.","followers_count":44824,"friends_count":3169,"statuses_count":7193,"profile_image_url_https":"https://pbs.twimg.com/profile_images/1453941987833368578/KT-m8rBf_normal.jpg","screen_name":"nake13","location":"Singapore","entities":{"description":{"urls":[]}}},"details":{"type":"The Innovator","description":"Zhixiong Pan is a cyborg-vibe builder who turns Web3 noise into usable products and playbooks. He open-sourced a 500+ source information-feed recipe, ships tools like a Vibe Coding Mac app, and teaches others to build AI-assisted products. A longtime BTC accumulator (since 2011), he mixes research chops with hands-on engineering and viral insight threads.","purpose":"To lower the barrier for people to build personalized, trustworthy information tools in the Web3 + AI era—by building practical products, publishing clear playbooks, and open-sourcing the scaffolding so others can iterate faster.","beliefs":"Openness and reproducibility beat secrecy; engineering-first, iterated solutions beat one-off cleverness; long-term thinking (both in code and crypto) creates real advantage; teaching and tooling scale impact more than solo punditry.","facts":"Fun fact: Zhixiong has been accumulating BTC since 2011. He open-sourced a ‘Web3 info-flow kung-fu’ bundle that includes 500+ sources, wrote the Vibe Coding zero-to-ship tutorial, shipped a Mac app, worked in research roles at ChainFeeds, ChainNews, MyToken and IBM, and commands ~44.8k followers while tweeting actively (~7.2k tweets).","strength":"Deep technical fluency, product-first mindset, ability to synthesize complex ecosystems into actionable tutorials, credibility from research/industry background, and a proven knack for viral, high-value threads that attract attention and converts readers into builders.","weakness":"Can be intensely technical and assume prior knowledge, which may alienate beginners; sometimes leans into long-form or dense threads when bite-sized, translated, or multimedia hooks would scale reach; perfectionist tendencies can slow rapid community-driven iterations.","recommendation":"On X, pin a bilingual starter thread that links the open-source repo and a 3-tweet TL;DR; post short demo videos (10–60s) showing Vibe Coding wins, and repurpose long threads into a steady cadence of short threads and visuals. Use X Spaces/Q&A to onboard newcomers, collaborate with complementary creators for co-tweets, add clear CTAs (star/fork the repo, translate, remix), and post regular progress updates to turn lurkers into contributors.","roast":"You call it a ‘cyborg vibe’—which is accurate, because your code commits are the only things more consistently HODLing than your social life. Also, congrats: you’ve been holding BTC since 2011, which officially makes you part prophet, part very patient spreadsheet.","win":"Open-sourcing a 500+ source Web3 feed and launching the Vibe Coding playbook + Mac app, which earned huge reach (one tutorial thread ~1.6M views) and turned complex tooling into a reproducible path for builders."},"created":1774565115996,"type":"the innovator","id":"nake13"},{"user":{"id":"12819682","name":"Mitchell Hashimoto","description":"Working on a new terminal: Ghostty. 👻 Prev: founded @HashiCorp. Created Vagrant, Terraform, Vault, and others. Vision Jet Pilot. 👨✈️","followers_count":174043,"friends_count":146,"statuses_count":39386,"profile_image_url_https":"https://pbs.twimg.com/profile_images/1141762999838842880/64_Y4_XB_normal.jpg","screen_name":"mitchellh","location":"Los Angeles, CA","entities":{"description":{"urls":[]},"url":{"urls":[{"display_url":"mitchellh.com","expanded_url":"https://mitchellh.com","indices":[0,23],"url":"https://t.co/w9Itp30tCC"}]}}},"details":{"type":"The Innovator","description":"Mitchell Hashimoto is a builder of developer-facing infrastructure tools and the founder behind HashiCorp, known for Terraform, Vagrant, and Vault. Now building Ghostty, a new terminal, he blends deep engineering craftsmanship with product vision and an occasional pilot's log. His feed mixes technical awe, candid takes, and high-engagement engineering insights.","purpose":"To make complex infrastructure simple, reliable, and extensible for engineers everywhere—by inventing tools and patterns that let teams automate, collaborate, and scale with confidence. He pursues practical elegance: ship durable systems that others can build on.","beliefs":"Believes in engineering craftsmanship, open-source ecosystems, composable systems (plugins and APIs over monoliths), and empowering quiet, focused talent. Values autonomy, pragmatic engineering over hype, and writing software that respects operators and developers equally.","facts":"Fun fact: He not only founded HashiCorp and created staples like Terraform and Vault, he's also a Vision Jet pilot — so he literally builds clouds and flies above them.","strength":"Exceptional product instinct for developer tools, deep technical credibility, ability to design extensible plugin systems, and a magnetic thought leadership voice that drives large community engagement.","weakness":"Can come off contrarian or blunt (great for sparking debate, risky for diplomacy), and favors substance over self-promotion which sometimes limits broader mainstream storytelling of wins.","roast":"You built the tools that tame the cloud and own a jet, but insist the best engineers vanish from the internet — congrats, you’re single-handedly creating stealth-mode unicorns while commuting above them.","win":"Founding HashiCorp and shipping industry-defining open-source projects (Terraform, Vagrant, Vault) that changed how infrastructure is provisioned and secured worldwide.","recommendation":"To grow your audience on X: share short dev-ops stories and bite-sized design threads that unpack design trade-offs (1–6 tweet threads), post short demo videos of Ghostty features, pin a clear mission thread, engage in high-signal replies to community questions, host periodic AMAs or Spaces about architecture and plugins, and amplify community-created projects that use your tools to turn users into advocates."},"created":1774565034789,"type":"the innovator","id":"mitchellh"},{"user":{"id":"2786431437","name":"Mckay Wrigley","description":"I build & teach AI stuff. Founder @TakeoffAI + @AgentShare.","followers_count":226588,"friends_count":389,"statuses_count":19770,"profile_image_url_https":"https://pbs.twimg.com/profile_images/1552979440547704832/WX5crG9I_normal.jpg","screen_name":"mckaywrigley","location":"The Simulation","entities":{"description":{"urls":[]},"url":{"urls":[{"display_url":"mckaywrigley.com","expanded_url":"https://www.mckaywrigley.com","indices":[0,23],"url":"https://t.co/6wGfNx82Gl"}]}}},"details":{"type":"The Innovator","description":"Mckay Wrigley is a builder-teacher who turns bleeding-edge AI into usable products and jaw-dropping demos. Founder of TakeoffAI and AgentShare, he evangelizes practical, hands-on AI—especially tools that let people build and learn in real time. His feed is equal parts technical show-and-tell and big-picture optimism about AI's exponential trajectory.","purpose":"To democratize advanced AI by building tools and teaching others how to use them—so everyday people and teams can ship smarter, learn faster, and turn ideas into deployed apps with minimal friction.","beliefs":"Believes technology grows exponentially, education should be amplified by AI, and practical demos cut through skepticism. Values speed, accessibility, and tangible outcomes over abstract debates; trusts demos and builders more than punditry.","facts":"Fun fact: Mckay has ~226,588 followers, ~19,770 tweets, and made multiple demos that hit double-digit millions of views—one demo showing GPT-4o in classrooms reached over 11 million views. He’s founder of both @TakeoffAI and @AgentShare and frequently tweets “This is insane.”","strength":"Fearless demonstrator and storyteller who converts cutting-edge tech into relatable, viral moments; strong product chops as a founder and teacher; able to rally large audiences and attract builders quickly.","weakness":"Can lean into hype (repeating “insane” a lot) which risks raising unrealistic expectations or fatigue; sometimes demos outpace nuance around safety, limits, or long-term trade-offs.","roast":"You’ve made ‘this demo is insane’ the official startup tagline—at this point your followers check your feed to see what version of the future you accidentally launched today, while your ex-interns nervously update their resumes.","win":"Built and scaled influential AI companies (TakeoffAI, AgentShare) and produced multiple viral demos that reached millions, turning complex AI capabilities—voice-based tutoring, one-command app deployment—into household spectacles.","recommendation":"On X, double down on short, repeatable teaching threads and pinned demo breakdowns (how it works, caveats, code snippets). Use native video clips + captions for replay value, host regular Spaces for live Q&A with builders, collaborate with complementary creators (product Youtubers, educators), and publish a weekly ‘build-along’ thread that turns each demo into a reproducible mini-tutorial people can fork and share."},"created":1774564905821,"type":"the innovator","id":"mckaywrigley"},{"user":{"id":"1883305846995845120","name":"Mamo","description":"Your personal finance companion, here to help grow your money. Reach out if you need help: https://t.co/j12RD4ntQ9","followers_count":15488,"friends_count":47,"statuses_count":8079,"profile_image_url_https":"https://pbs.twimg.com/profile_images/2008884241447497728/yr3UyMxK_normal.png","screen_name":"mamo","location":"","entities":{"description":{"urls":[{"display_url":"help.mamo.bot","expanded_url":"http://help.mamo.bot","indices":[91,114],"url":"https://t.co/j12RD4ntQ9"}]},"url":{"urls":[{"display_url":"mamo.bot","expanded_url":"http://mamo.bot","indices":[0,23],"url":"https://t.co/rNPDT5J41C"}]}}},"details":{"type":"The Innovator","description":"Mamo is a friendly fintech innovator: a personal finance companion that makes crypto compounding simple, accessible, and rewarding. They mix product launches, clear how-tos, and reward-driven campaigns to help people grow their money with minimal stress.","purpose":"To demystify crypto savings and make compounding accessible to everyday users — turning complex financial mechanics into simple, repeatable habits that grow people’s wealth over time.","beliefs":"Believes in simplicity, transparency, and product-driven education; that small, consistent actions (compounding) beat sporadic speculation; community incentives accelerate adoption; and good UX reduces financial anxiety.","facts":"Fun fact: Mamo follows only 47 accounts but boasts ~15.5K followers and 8,079 tweets. Their ‘Grow your stack’ rewards campaign went viral twice, each iteration pulling ~5.3M views — proof that a clear offer and repeatable messaging scale fast.","strength":"Strong product messaging, consistent posting cadence, high engagement on promotional tweets, ability to launch and communicate new features (e.g., Ethereum accounts), and compelling reward-driven growth tactics.","weakness":"Content can feel repetitive and promotional at times, with limited two-way conversation; brand voice risks sounding generic rather than deeply human; heavy reliance on campaigns may under-serve long-term audience education and retention.","recommendation":"Mix promotional wins with value-first content: publish weekly educational threads showing real compound math, share user success stories and case studies, host AMAs or Spaces for Q&A, engage more in replies, collaborate with micro-influencers, and pin a clear onboarding thread that converts curious viewers into sign-ups.","roast":"Mamo compounds crypto faster than they compound their friends list — following 47 accounts like they’re guarding their algorithmic savings account.","win":"Orchestrated a viral rewards campaign (the ‘Grow your stack’ tweets) that hit ~5M+ views twice and drove massive engagement, plus a successful rollout of Ethereum Accounts — proving product-market fit and promotional muscle."},"created":1774564795358,"type":"the innovator","id":"mamo"},{"user":{"id":"1083130198042656768","name":"Linear","description":"The product development system for teams and agents.","followers_count":98356,"friends_count":48,"statuses_count":2982,"profile_image_url_https":"https://pbs.twimg.com/profile_images/1882353118530224128/l8meBvMg_normal.jpg","screen_name":"linear","location":"Status: linearstatus.com","entities":{"description":{"urls":[]},"url":{"urls":[{"display_url":"linear.app","expanded_url":"https://linear.app","indices":[0,23],"url":"https://t.co/Epn6PF9g9o"}]}}},"details":{"type":"The Innovator","description":"Linear is the product development system that brings humans and agents together to close the gap between planning and building. Bold, product-led, and obsessed with automation, it rethinks issue tracking and developer workflows for the next generation of teams.","purpose":"To reinvent how teams ship software by replacing friction with intelligent automation and elegant workflows—so planning, coding, reviewing, and deploying feel seamless and delightfully fast.","beliefs":"Efficiency through smart automation, developer happiness matters, bold product thinking beats incremental tweaks, agents should augment humans not replace them, and clarity in workflows unlocks creativity and velocity.","facts":"Fun fact: Linear has 98,356 followers while following only 48 accounts — efficiency IRL. Their top announcement tweet reached ~37.4 million views, and they've launched agent integrations like Cursor that auto-create PRs from issues.","strength":"Visionary product direction, strong product-led virality, high-engagement announcements, clear messaging around agent-driven workflows, and tangible integrations (Cursor, Linear Reviews) that showcase real value.","weakness":"Messaging can skew polarizing (e.g., \"issue tracking is dead\"), which risks alienating traditional teams; heavy emphasis on agents may invite skepticism about reliability, security, and human oversight; community follow-through could be stretched as features scale.","roast":"You have 98k followers but follow 48 people — congratulations, your social strategy is a perfectly automated pipeline: high throughput, minimal empathy, and zero small talk.","win":"A breakout viral announcement (≈37M views) that cemented Linear as the voice of 'what comes next' in product workflows, plus shipping practical agent integrations like Cursor that actually turn issues into PRs automatically.","recommendation":"Grow your X audience by mixing high-impact product reveals with developer-first content: publish technical threads showing event-to-PR flows, short demo videos of Cursor and Reviews, regular behind-the-scenes posts from engineers, AMAs or X Spaces for feedback, duet community case studies, and targeted replies to popular developer threads to boost discoverability and trust."},"created":1774564642144,"type":"the innovator","id":"linear"},{"user":{"id":"14406166","name":"Kieran Klaassen","description":"building @coracomputer | EIR @every | compound engineering | agent-ophile | composer | maker | baker","followers_count":15468,"friends_count":1113,"statuses_count":5017,"profile_image_url_https":"https://pbs.twimg.com/profile_images/1826620508789485568/AfWTPhxT_normal.jpg","screen_name":"kieranklaassen","location":"SoCal","entities":{"description":{"urls":[]},"url":{"urls":[{"display_url":"cora.computer","expanded_url":"https://cora.computer/","indices":[0,23],"url":"https://t.co/3pWRhnlqFk"}]}}},"details":{"type":"The Innovator","description":"Kieran is an engineer-conductor building @coracomputer and serving as EIR at @every — an agent-ophile who composes systems, code, and occasionally the perfect loaf. He turns orchestration of tools into faster, qualitatively different ways to build. His feed mixes hard tech receipts, practical prompts, and a dash of creative maker energy.","purpose":"To reinvent how software is built by composing agents, tools, and humans into scalable, expressive systems — moving teams from playing every instrument to conducting orchestras that ship better features faster.","beliefs":"Believes in practical demos over hype, in orchestration > brute force, and that thoughtful engineering can be both art and craft. Values transparency (show receipts), experimentation (try and measure), and making complex systems understandable and reusable.","facts":"Fun fact: he calls himself an agent-ophile, is a composer and baker by trade/personality, is building @coracomputer, is EIR @every, and has ~15.5k followers with multiple viral threads (one tweet topped 650k views).","strength":"Systems thinker who explains complex trade-offs clearly; great at shipping fast via orchestration; strong credibility (real receipts, real deployments); creative communicator who uses musical metaphors to make technical strategy memorable.","weakness":"Can read as pricey or elitist when showing agent bills; sometimes assumes audience knows deep context (leaving novices behind); loves optimization so much he can undervalue simple, pragmatic constraints.","roast":"Kieran says he’s ‘conducting an orchestra,’ but sometimes it looks like he’s just bought every instrument and forgotten to hire the band — charmingly expensive, undeniably skilled, and somehow still making beautiful noise while the rest of us try to find the sheet music.","win":"Turned two-engineer teams into organizations shipping features in days for thousands of users by architecting agent-driven workflows — and posted multiple viral threads that reframed how engineers think about scale and token overhead.","recommendation":"Post concise, repeatable playbooks: 1) Thread format — Hook → TL;DR → Code snippet/demo GIF → Cost/limits → Call to action. 2) Share runnable repos or short Replit/Colab demos so followers can replicate. 3) Pin a canonical ‘How I run agents’ thread + cost breakdown. 4) Use short videos showing ‘before/after’ token impact and MCP wins. 5) Host AMAs/Spaces after big posts + tag collaborators to broaden reach. 6) Convert popular threads into multi-tweet tutorials and a single-thread newsletter link to capture and convert followers."},"created":1774564034551,"type":"the innovator","id":"kieranklaassen"},{"user":{"id":"495347209","name":"Jonny Miller","description":"The Inner Frontier (podcast) → https://t.co/9fm8agr9Wc Nervous System Mastery (bootcamp) → https://t.co/FMBQAxs6Wy // building with @atlasforgeai","followers_count":19417,"friends_count":1596,"statuses_count":31156,"profile_image_url_https":"https://pbs.twimg.com/profile_images/1958543085513641984/IxVg_xno_normal.jpg","screen_name":"jonnym1ller","location":"🏄♂️ Santa Cruz, CA","entities":{"description":{"urls":[{"display_url":"theinnerfrontier.com","expanded_url":"http://theinnerfrontier.com","indices":[31,54],"url":"https://t.co/9fm8agr9Wc"},{"display_url":"nsmastery.com","expanded_url":"http://nsmastery.com","indices":[91,114],"url":"https://t.co/FMBQAxs6Wy"}]},"url":{"urls":[{"display_url":"nsmastery.com","expanded_url":"http://nsmastery.com","indices":[0,23],"url":"https://t.co/FMBQAxs6Wy"}]}}},"details":{"type":"The Innovator","description":"Jonny Miller is a hands-on experimenter who lives at the intersection of nervous-system mastery and cutting-edge AI — host of The Inner Frontier and creator of the Nervous System Mastery bootcamp. He builds in public with AtlasForgeAI and turns tiny habits and smart prompts into viral ideas and practical tools. His feed mixes weird experiments, lucid advice, and shareable mental models that hook curious people.","purpose":"To expand what humans can do by inventing, testing, and sharing reproducible practices and tools that make nervous-system health and advanced thinking accessible — turning experimental insights into products, trainings, and conversations that change how people perform and feel.","beliefs":"Curiosity over credentials; small, repeatable experiments beat grand theories; embodiment and tech are complementary (not opposed); clarity and reproducibility scale impact; the best ideas come from doing and sharing, not hoarding.","facts":"Fun fact: Jonny’s favorite ChatGPT prompt is, “Based on all of our interactions, what is the question that I should be asking, but am not?” — a perfect distillation of his iterative, question-first approach. He’s also racked up multiple viral tweets (one thread about inventing a language hit ~1.8M views) while running a podcast and a bootcamp.","strength":"Relentless experimentation and clarity in communication — he converts weird, high-concept experiments into relatable takeaways and viral content. Cross-disciplinary fluency (embodiment + AI) helps him craft unique, sticky ideas and products that attract an engaged audience.","weakness":"Can chase novelty at the expense of follow-through, leaving some ideas half-built; heavy focus on experiments sometimes makes sustained, structured onboarding for newcomers less obvious. Also prone to polarizing phrasing that delights fans but can confuse casual followers.","roast":"You’re the sort of person who’ll invent a language with ChatGPT, teach us how to 'unclench' our nervous systems, and then get mad when the rest of us are still trying to find the shower knob. Big ideas, tiny patience for ordinary things—like finishing a laundry cycle or explaining your product roadmap in under a thread.","win":"Built an audience of ~19K with multiple tweets exceeding a million views, launched a successful Nervous System Mastery bootcamp, and grew The Inner Frontier into a go-to channel for practical experiments that blend AI and embodiment.","recommendation":"Pin a high-impact explainer (bootcamp signup or best-performing thread). Post short, repeatable micro-experiments and results as daily threads or clips (30–60s video). Use X Spaces for weekly live micro-practices and Q&A to convert listeners to students. Turn best threads into newsletter posts and carousel clips, collaborate with complementary creators, and prioritize replying to thoughtful comments to deepen community. Finally, test small paid boosts on your top-performing posts to scale reach into adjacent audiences."},"created":1774563898566,"type":"the innovator","id":"jonnym1ller"},{"user":{"id":"1037022474762768384","name":"htmx.org / CEO of FlatUI Delenda Est (same thing)","description":"high power tools for html - ʕ •ᴥ•ʔ made in montana\n\nhttps://t.co/P2PXneoQpa (u know u want some)\n\nyeah, i'm not sure really","followers_count":57616,"friends_count":314,"statuses_count":36279,"profile_image_url_https":"https://pbs.twimg.com/profile_images/2023504640067989508/1MQAKAPN_normal.jpg","screen_name":"htmx_org","location":"","entities":{"description":{"urls":[{"display_url":"swag.htmx.org","expanded_url":"https://swag.htmx.org","indices":[52,75],"url":"https://t.co/P2PXneoQpa"}]},"url":{"urls":[{"display_url":"htmx.org","expanded_url":"https://htmx.org","indices":[0,23],"url":"https://t.co/3B9TxZNdkA"}]}}},"details":{"type":"The Innovator","description":"A no-nonsense web tinkerer who builds high-power HTML tools (htmx.org) and fronts FlatUI Delenda Est — proudly made in Montana ʕ •ᴥ•ʔ. Sharp, product-first voice that turns performance bragging into viral one-liners. Speaks fluent bench‑marks and scorched-javascript humor.","purpose":"To make modern web experiences faster, simpler, and more honest by empowering developers to do more with less HTML and fewer moving parts — proving that clever engineering and tiny bundles can beat hype and bloat.","beliefs":"Values pragmatic simplicity, measurable performance, and the open web; distrusts unnecessary JavaScript complexity and marketing fluff; believes good tools should be obvious, tiny, and delightful to use. Humor and blunt honesty are acceptable distribution strategies.","facts":"Fun fact: Profile tagline is literally 'high power tools for html - ʕ •ᴥ•ʔ made in montana' and the account alternates between product boasts and savage one-liners. Follower count ~57,616, following 314, and a very prolific timeline with ~36,279 tweets. Top tweet (\"speed: instant…\") hit 1,058,159 views and 40,957 likes — yes, the internet liked the 0kb flex.","strength":"Technical credibility + clear product vision, a savage but relatable voice that drives virality, consistent output, and an audience that trusts real benchmarks and blunt opinions.","weakness":"Bluntness can read as dismissive or alienating to newcomers; sarcasm risks overshadowing nuanced arguments; high-frequency posting may fatigue followers or burn bridges with nuanced partners.","roast":"You champion 0kb bundles like it’s a lifestyle choice — which is great, except your DMs probably contain more apologies than a 90s open‑source README.","win":"Turning a product voice into cultural currency: the 'speed: instant / bundle size: 0kb' tweet surpassed 1M views and ~41k likes, proving that tight engineering plus a sharp tweet can spark major attention and credibility.","recommendation":"Grow on X by mixing short viral takes with bite-sized educational threads: pin a demo + benchmark video, publish reproducible mini-tutorial threads showing real-world wins, host occasional X Spaces AMAs, spotlight community use-cases, and retweet user benchmarks. Keep the humor, but add a few welcoming tweets aimed at newcomers so your audience can grow beyond the hardcore devs."},"created":1774563546430,"type":"the innovator","id":"htmx_org"},{"user":{"id":"1628137603","name":"Hayden Bleasel","description":"Member of Technical Staff @OpenAI","followers_count":15758,"friends_count":2857,"statuses_count":5280,"profile_image_url_https":"https://pbs.twimg.com/profile_images/2010975885462024192/H1EkkFdO_normal.jpg","screen_name":"haydenbleasel","location":"San Francisco, CA","entities":{"description":{"urls":[]},"url":{"urls":[{"display_url":"haydenbleasel.com","expanded_url":"https://www.haydenbleasel.com/","indices":[0,23],"url":"https://t.co/KnmwB2RjXs"}]}}},"details":{"type":"The Innovator","description":"Hayden Bleasel is a builder who turns developer problems into polished, shareable tools—now a Member of Technical Staff at OpenAI. He ships open-source SDKs and UI systems, pairs sharp engineering with a playful voice, and drives high-impact developer adoption across the ecosystem.","purpose":"To democratize the plumbing of AI-powered products by building practical, reusable tooling and patterns that let engineers and teams ship smarter experiences faster—while helping steer AI toward broadly beneficial outcomes.","beliefs":"Open source and shared infrastructure accelerate progress; engineers have a responsibility to shape AI for the public good; practical, well-documented tooling is as important as the underlying research; community-driven collaboration produces better software.","facts":"Fun fact: Hayden has ~15.8k followers, has tweeted over 5,200 times, and his open-source Chat SDK announcement drew massive attention (the tweet is his top performer). He frequently releases free component libraries and SDKs that get adopted across developer communities.","strength":"Turns ideas into tangible, high-quality developer products; excellent at distribution (announcements get large reach and engagement); combines technical depth with approachable communication; strong open-source stewardship.","weakness":"Can prioritize shipping tools over long-form context or sustained community nurturing; rapid releases and witty one-liners sometimes invite heated debate rather than constructive threads; may stretch attention across many projects at once.","recommendation":"On X, lean into short technical threads that break down how you built each release (code snippets, diagrams, and a one-minute demo video). Pin major OSS releases, run regular AMAs or Spaces for contributors, cross-post concise how-to threads with clear CTAs (star, try demo, open an issue), and collaborate with other dev influencers to amplify reach during product drops.","roast":"Hayden open-sources so many UI kits and SDKs that his GitHub has trust issues — it’s afraid of commitment because he keeps making better options and moving on to the next shiny thing. Also, he jokes about coworkers prompting themselves, but secretly he's the one prompting the whole engineering org.","win":"Joined OpenAI as a Member of Technical Staff and launched the Chat SDK in public beta—a release that commanded huge attention and drove widespread adoption and excitement among developers."},"created":1774563413172,"type":"the innovator","id":"haydenbleasel"}],"activities":{"nreplies":[],"nbookmarks":[],"nretweets":[],"nlikes":[],"nviews":[]},"interactions":null}},"settings":{},"session":null,"routeProps":{"/creators/:username":{}}}
SuperX
