Get live statistics and analysis of Lucian in sf's profile on X / Twitter

I write about tech, future, mind, health. built maginary.ai, a midjourney-like img/vid generator that reads/blows your mind

630 following595 followers

The Innovator

Lucian in sf is a forward-thinking tech enthusiast who blends mind-expanding ideas with cutting-edge digital creativity. They are not just dreaming about the future—they're building it, as seen in their work with a mind-blowing image and video generator. Lucian sparks curiosity and challenges norms with bold, philosophical takes on biology, society, and technology.

Impressions
222.1k-26.4k
$41.64
Likes
169-32
81%
Retweets
3
1%
Replies
30-9
14%
Bookmarks
61
3%

Top users who interacted with Lucian in sf over the last 14 days

@oranahhX

Building Apps to fund a global animal rescue movement 🎯

3 interactions
1 interactions
@cyber_molecule

Founder | Write about Product Strategy and Product Market Fit for SaaS and Startups | Building mapster.io/?ref=xp

1 interactions
1 interactions
@buildItN0w

Code hard 💻 Ship fast 🚀 Sleep later 😴 🧠 ShowUpInAI.com 📸 PostCapture.com 🐶 BacklinkDog.com 🥳 ConfettiSaaS.com

1 interactions
@AngelicaIAC

CEO @nucleo_research (YC F25) - AI x Oncology👩‍🔬- Ex @Stanford @IITMadras

1 interactions

Lucian’s tweets are like that one friend who answers the simplest question with a TED talk — brilliant but sometimes you just wanted a yes or no. Maybe toss in a meme once in a while to remind people you’re human and not a wandering AI algorithm yourself.

Successfully built and launched a user-friendly AI image and video generator with an open-access model, demonstrating technical prowess and a commitment to democratizing creative technology.

To push the boundaries of technology and human potential by creating tools and ideas that empower people to transcend limitations and rethink their reality.

Lucian believes in decentralized cooperation, the power of self-mastery, and that traditional social constructs like countries are artificial barriers to human harmony. They value freedom, curiosity, and challenging the status quo with a vision for a better, more interconnected world.

Lucian excels at merging deep philosophical insights with practical tech innovation, inspiring others to rethink their limits while providing tangible, advanced tools to fuel creativity and growth.

Their complex, visionary ideas can sometimes be too abstract or unconventional for broader audiences, which may limit immediate engagement or understanding.

On X, Lucian should leverage thread storytelling to break down complex topics into bite-sized, relatable insights and demonstrate ongoing projects. Engaging visuals and collaborations with other creators could amplify reach and foster a passionate following.

Lucian has developed a midjourney-like AI-powered image and video generator that accepts any language and format, with no forced subscriptions—an impressive feat showcasing their innovative spirit.

Top tweets of Lucian in sf

we are all playing a multiplayer game with biology's non-authoritative server, meaning each of us have the whole game rules in our own dna, more or less the same. it's more like a peer to peer game at the gene level that means your dna decides when you should lose in order to let others win, because those are the game's rules when a male-goat loses a battle, its testicles literally shrink and it becomes infertile for the year, and the winning one gets more confident and his T increases to the roof. but there's no central server commanding this -- there has to be cooperation at the gene level, independently and decentralized when you get rejected (job, dating, social), you might enter a permanent self reinforcing doom loop, self-doubt, self-deprecation and actually manifest how you THINK those who rejected you see you you know what else you can do in a peer to peer game? you can cheat like crazy, because, again, your dna/brain has all the rules and tries to enforce them on you, but you, knowing that, can just modify the the source code/configs the way you want. obviously you can't grow wings, but you can bypass MANY more limitations than you think -- you can, for instance, ignore how you feel about anything negative in life. not try to deny it, just realize it's your version of the game that tries to confirm someone ELSE has won (in a certain situation), because it has to sync with what others' dnas think somewhere in this world there's a guy who got rejected by a girl in highschool and has lived his whole life on a downward trajectory, knowingly or unknowingly, maybe not even talking about it, but its effects flowing through all his life's cracks, amplifying misery. many such cases nietzsche is the goat, the winning goat every boundary is psychological you are the master if your life you can have everything you want go and take it. don't ask for permission just don't harm others, that's all

19k

introducing rockstart gamedev a live #gamedev course for my free edu initiative this is an intensive 4-week deep dive into the unity game engine, but also in a game developer's life, a to z (this is the first course on the platform, setting a standard for quality > quantity) - go to rockstart's site to enter the discord - go to the yt channel and start watching the course - when ready to take the tests, dm me here/discord until the rockstart platform will be ready - win! went from 1 to 53 subs during this time, 5 people on average joined in these 3 weeks of livestreams, but hey! the course is available for free and you can enroll at any time and win a certificate -- all for free! built during @_buildspace s5 thanks @FarzaTV @jeffrey_notes & the team for all your knowledge -- really, this is not just me being polite, you guys have the highest signal-to-noise ratio I ever saw

1k

day 24 I'm working on doc2exam, a place to turn any material into live exams -- for students prepping or professors setting official certification today I improved llamaindex's own prompts for keyword generation (basically created my own), to be much more accurate, and got more familiarized with structured outputs, particularly with the Ollama llm in llamaindex apart from explicitly asking for json output and including an output example in the prompt, I had to turn off function calling, which is on by default and llamaindex complains about it, and set json mode=True I realized I shouldn't do all the clustering at the doc ingestion time -- only keyword generation per node (which is now pretty good btw) this is because user can have a doc attached to many exams and preemptively tracking multiple clusters for that via metadata in vector db seemed like overkill instead, when the user actually starts generating questions, I cluster all nodes related to the attached docs only and generating subjects from the keywords in those clusters all in-memory (for now). keeps it simple until I really need scaling. and also adding a cache later will speed subsequent question generations even more, so I have an easy path to optimizing I take note of keyword frequency when generating subjects, and then of subject frequency when generating questions, so that I can cover the materials uniformly while avoiding repeating existing questions. the question generation itself is on today's menu (I thought I'd be done with it by now) this will use llama3.1:70b on @huggingface via inference client (I have a sub with them) I don't think this will scale past 100 users, as hf severely rate-limits concurrent requests, even with a pro account, at least in my case, and they have 10k requests per day if a user presses the "Generate" button 30 times per day on average, that's 333 users, let's say 300, but I'm reusing that hf sub for another project, so I think 100 users is a decent target for now -- ofc even 10 users will make me really happy in the launch week 😂 I also fixed some segfault error in celery due to a bug in their multiprocessing support. so to get true parallelism on the same machine I'll have to spawn multiple workers -- a problem for later. because of that, I also got slidetracked for an hour or so with looking into alternatives: - dramatiq has unimaginative licensing (LGPL) - huey is pretty much the same, probably better but not enough to justify another learning curve, no matter how gently - taskiq is promising as it has native asyncio so might be next choice once I find pressing reasons to switch from celery work continues! day 23: x.com/xucian_/status…

137

day 11 I'm working on doc2exam, a place to turn any material into live exams -- for students prepping or professors setting official certifications a special day of the year, so I filled it with with a copy of The Lean Startup. so while it's not work, it definitely recharges you in this direction. I'm strongly against just building each day -- that deep work is the main thing you should do, but to avoid wasteful local minima you have to zoom out once in a while speaking of recharging, I'm watching this daily: x.com/0xluffyb/statu… while not an acolyte of the e/acc movement (I'm neutral, cults make me question my legitimacy), the video is really good (and I really want to create one myself soon, I think it'll be incredible, unbranded, i.e. without all the implied "let's do things as fast as possible, because, nothing could go wrong") not that I'm against e/acc, I just prefer to delay my judgement on it and just encourage people to simply build stuff.. driven not by frenzy, but by excitement and for the runners reading this, I posted an incredibly good indie hacker-friendly mix in comments alright fellas, back to building the future! day 10: x.com/xucian_/status…

126

rockstart #gamedev course with #unity today's live-coding session [2/3] for the Duck Invaders game was also nice! we've created the wave system (used a simple variation of an FSM again), including the waves for lvl 1, hit effects, we've created more enemy types, added realistic stars scrolling in the background, and tried to generate a soundtrack with musicgen (forgot to come back to it, but it was just 15secs anyway, so tomorrow we'll generate one locally, which can be up to few mins) below the end result! see you tomorrow for the next live-coding session! [3/3] w3d3 linked below, as well as tomorrow's scheduled stream! CC: @_buildspace @_nightsweekends @FarzaTV

107

rockstart #gamedev course with #unity a video recap of week2 shorter text version: x.com/xucian_/status… CC: @_buildspace @_nightsweekends @FarzaTV

65

Most engaged tweets of Lucian in sf

we are all playing a multiplayer game with biology's non-authoritative server, meaning each of us have the whole game rules in our own dna, more or less the same. it's more like a peer to peer game at the gene level that means your dna decides when you should lose in order to let others win, because those are the game's rules when a male-goat loses a battle, its testicles literally shrink and it becomes infertile for the year, and the winning one gets more confident and his T increases to the roof. but there's no central server commanding this -- there has to be cooperation at the gene level, independently and decentralized when you get rejected (job, dating, social), you might enter a permanent self reinforcing doom loop, self-doubt, self-deprecation and actually manifest how you THINK those who rejected you see you you know what else you can do in a peer to peer game? you can cheat like crazy, because, again, your dna/brain has all the rules and tries to enforce them on you, but you, knowing that, can just modify the the source code/configs the way you want. obviously you can't grow wings, but you can bypass MANY more limitations than you think -- you can, for instance, ignore how you feel about anything negative in life. not try to deny it, just realize it's your version of the game that tries to confirm someone ELSE has won (in a certain situation), because it has to sync with what others' dnas think somewhere in this world there's a guy who got rejected by a girl in highschool and has lived his whole life on a downward trajectory, knowingly or unknowingly, maybe not even talking about it, but its effects flowing through all his life's cracks, amplifying misery. many such cases nietzsche is the goat, the winning goat every boundary is psychological you are the master if your life you can have everything you want go and take it. don't ask for permission just don't harm others, that's all

19k

wow just had a chat with @bookingcom's chatbot regarding an invoice for my flight, I think it was the fastest, most useful, least BS chatbot convo and resolution I've seen in my life, congrats bros

96

finishing slides for the first #unity #gamedev course for rockstart.ai more info: x.com/xucian_/status… few slots still available! will post updates here

184

hello everyone! 8y ago I was launching Optimized ScrollView Adapter, a UI plugin for the #unity game engine I just compiled everything about the journey in under 3 mins! if you're a #gamedev and have used it, please comment below! means a lot

67

rockstart #gamedev course with #unity today's live-coding session [3/4] for the Duck Invaders game also took 4h! yeah, I decided to do 4 of these, so including tomorrow! we've generated the soundtrack with ai, added lots of cool sound effects from @pixabay (with attribution!), and even using the mixer to blend the soundtrack with the sfxes like pros do! created a new weapon "AoE", ducks can now drop weapon upgrades randomly, but they also lie eggs (that kill you! this is really something imo, *pat on the back* see you tomorrow for the next live-coding session! [4/4] w3d4 linked below, as well as tomorrow's scheduled stream! CC: @_buildspace @_nightsweekends @FarzaTV

85

rockstart #gamedev course with #unity today's live-coding session [2/3] for the Duck Invaders game was also nice! we've created the wave system (used a simple variation of an FSM again), including the waves for lvl 1, hit effects, we've created more enemy types, added realistic stars scrolling in the background, and tried to generate a soundtrack with musicgen (forgot to come back to it, but it was just 15secs anyway, so tomorrow we'll generate one locally, which can be up to few mins) below the end result! see you tomorrow for the next live-coding session! [3/3] w3d3 linked below, as well as tomorrow's scheduled stream! CC: @_buildspace @_nightsweekends @FarzaTV

107

day 12 I'm working on doc2exam, a place to turn any material into live exams -- for students prepping or professors setting official certifications today (Wednesday) I played with the LLM/RAG pipeline, it's incredible how each project starts by underestimating its complexity -- sometimes, knowing too much would prevent you from starting something 😂 first of all, as we all know, LLM's output is non-deterministic. in my case, I want the exam's questions to be generated as json, so I dug the internet and found the following python libs (links in comments): 1. outlines and guidance -- they look really promising as they're restricting the generation itself at the token-level (and arguably this decreases accuracy, but I think not by much, so it looks like a net positive). problem is, I can't use them because it seems there's no way to hook it together with @huggingface's InferenceClient, i.e. I'd have to use a local model (which I plan to, but only for the small stuff, like classification) 2. json-repair -- which basically post-process your prompts. it's not ideal to only rely on this, but it's a huge upgrade. seems to work better than my own, so this is what I'm using. it also removes LLM's comments such as "Here's the updated json" on top of that, I discovered hf's InferenceClient has a response_format param where you can send a json schema (hint: if you have pydantic models, you can use their model_json_schema function to generate it), looks really promising, I'm using it now but didn't fully test it yet, and there's little documentation about it then also went onto processing the materials themselves. this is wip, but generally I'm segmenting them into chapters, see what questions already exist in order to skip those chapters, then decide which chapters to use for the generation, extract them via RAG and send them in the prompt other than that, I improved the front-end navigation bar where now it highlights the current page this is the product's core, it has to be resilient and deterministic. so back to work! day 11: x.com/xucian_/status…

89

day 15 I'm working on doc2exam, a place to turn any material into live exams -- for students prepping or professors setting official certifications yesterday (Saturday) I finally got my server laptop to run and serve ollama (settled with linux mint). I spent some time checking ollama's flash attention env var, but I haven't found any speedups for llama3.1 70b, nor 8b. if anyone reading this knows why, lmk. logs showed "flash_attn = 1" so it was passed. anyway, not a core necessity rn the models usage will probably be something like: - exam question generation: llama3.1:70b via huggingface's inference api (my local 70b on rtx30708gb is still pretty slow) - student's answer evaluation (1 by 1): llama3.1:8b via ollama - rag/embeddings: (some model) via ollama - summarizations: llama3.2:3b via ollama this should save me tons of api calls, but will also be more predictable and no stupid rate limits even if I mess up an api call in an infinite loop 😆 linux mint feels really nice, and stuff works pretty much out of the box (incl nvidia drivers). I still love debian, but will stick to it on my servers only btw, added a video just for fun with my local llama70b struggling with outputing 1.0 (the score for a perfect/near-perfect answer). then switched to 8b for comparison (1st try includes models initial loading, so 2nd try at 0:50 is the effective speed) what's next? mapping the above models into the whole llm pipeline (and still deciding how much of each of llamaindex/dsrag/langchain to use. I know the answer is simple for who already knows it, but the paradox of choice always pushes me up the midwit peak at first😅) day 14: x.com/xucian_/status…

229

People with Innovator archetype

The Innovator

DevRel Engineer @cloudflare 🥑 • Web Platform Advocate 🌐 • Learning Rust 🦀 • Software Engineer 🛠️ • 鼓舞する ✨ • confidence.sh/blog

213 following2k followers
The Innovator

Crypto, Web3, Blockchain Designer :)

1k following1k followers
The Innovator

15 y/o | Web3 and Rust enthusiast | Full Stack developer | Aot lover

176 following106 followers
The Innovator

MTS at @FactoryAI; previously at: Meta, Airbnb, Brex & Opendoor

331 following647 followers
The Innovator

jesus is king. i push buttons, things unbreak. tse @docker

476 following137 followers
The Innovator

Product Engineer, Nomadism, Hodler of Stable Coin&Crypto ETFs. Co-Founder of @ibuidl_2024 & @Nomoslabs, ex-@nnsdaos

1k following673 followers
The Innovator

Building quietly, acting loudly. Crypto, AI, and chaos — I connect the dots before they make sense. Here for innovation.#kaito #Bio #DeFi #AI #Web3❤️❤️❤️

4k following2k followers
The Innovator
67 following124 followers
The Innovator

✨ Web3 Storyteller ✨ From AI-driven insights to community-driven memes. Following the pulse of @KaitoAI // @cookiedotfun // @bioprotocol

601 following374 followers
The Innovator

Cryptocurrency lover

2k following680 followers
The Innovator

Researching new optimal ways in web3 marketing. Also spreading my views here

1k following1k followers
The Innovator

A curious mind in the world of DeFi Always one step ahead with Kaito 🚀

2k following1k followers

Explore Related Archetypes

If you enjoy the innovator profiles, you might also like these personality types:

Supercharge your 𝕏 game,
Grow with SuperX!

Get Started for Free