Get live statistics and analysis of Roman's profile on X / Twitter

If you think the AI space is too crowded Here’s a goldmine nobody’s talking about: Phone automations. iOS 26 (dropping this fall) is launching full-blown Automations, basically a mini n8n built into your phone. It’s not just Shortcuts anymore. This is native, trigger based, real world automation for normies. And nobody’s ready. It will feature triggers, like n8n but much cooler: - when you connect to Wi-Fi or Bluetooth - when opening a specific app - when sending a text - when switching focus modes - when pluged into your car - when calling your mom (literally) And here’s where it gets wild You can send/receive webhooks Which means? You can chain your phone to n8n and back. Offload the heavy lifting to real backends, then return outputs straight to your pocket. This isn’t a party trick. It’s a playground for building consumer AI products that feel native, magical, and stupidly sticky. First things that come to mind: – GPT car copilots (triggered by Bluetooth in your Toyota) – productivity bots that respond to location, weather, or calendar – “focus mode” agents that coach, hype, or guide your day – instant journaling, expense tracking, meal logging, outfit planning—whatever – voice controlled workflows with no app download, just an AirDrop install $29.99/month wrappers ready to print. So what happens next? You can wait till fall, Do fucking nothing, goon over IG motivation videos Buy your new iphone, when it comes out, like a fucking mr consumer And goof around, play dress up barbie with your phone changing wallpapers OR You can learn how "Shortcuts" work at it's current state Go balls deep into automations, when it comes out during public beta in july Learn how make it "collab" with n8n Research consumer apps and learn what could be potentially selling Stack knowledge & prep plug-n-play automations This is your shot to build in a niche before it gets named. Don’t wait for launch hype. Dig your trenches now. Because this fall? You’re either cashing in or updating your lock screen.

as time goes on, you start seeing it clear there’s people who consume new prompt, new yt video, new AI hack watch this, scroll that, they feel busy but 0 gets done and there’s people who just build no bs, no fakery, no fluff the internet’s tricky like that flashy posts, filtered wins, curated success but here’s the thing looking successful in front of strangers will never beat actually being successful in real life especially when the people clapping don’t even know your name

“Oh wow iOS 26 🥹” while everyone’s busy licking Apple’s liquid ass, the real update just flew over their heads native automations no shortcuts. no raindance. no code. best part? you can rip out Apple’s limp "intelligence" and plug in ChatGPT like it’s a nokia battery upgrade natively you now can run automations right from your damn phone and what are most people doing? playing dress-up, changing lock screen use all brain cells hook it up to the beloved n8n automate like a psychopath airdrop it to a client like you're on Apple’s payroll who's doing this?

I just sent a website draft to a client for review But instead of making them send emails, screenshots, or voice notes full of "change that blue thing near the what’s it called" I gave them a little chatbot. Right there on the website Built in Voiceflow. Very basic prompt Took like 20 minutes to set up. Client clicks through the site, talks to the bot as if they’re texting me “Move this button up” “Make this font bolder” “Can we try a darker green?” The bot logs everything into Airtable in real time. No back and forth No deciphering vague emails No screenshots with red circles Just clean, structured feedback Straight into my build pipeline. Saves me time Saves them effort Feels like a game Best part? Feedback feels “fun” instead of being perceived as "extra shit I need to do" Not a flashy agent Just a small, boring automation in the exact right place.

You’ve seen it a hundred times by now: DeepSeek r1 – 671B parameters GPT-4o – 1 trillion parameters Okay but what the hell are these parameters? Is this a who has a bigger d*ck competition? When people say a model has “1 trillion parameters,” they’re talking about how many tiny little knobs it has under the hood. These knobs are what the AI tweaks during training to get better at predicting text, solving problems, or just sounding smart. The more knobs it has, the more it can learn. Simple as that. Technically, these knobs are just numbers, weights and biases in a neural network. Every time the model sees a sentence, it’s adjusting those numbers. “This word should matter more, that one less. Oh, and let’s not forget this comma.” Over billions of examples, it gets good at spotting patterns, grammar, logic, tone, everything. So when it sees “I love” it knows “you” “this” are good guesses, but “pee” or “murder” might raise an eyebrow. The transformer architecture (what all modern LLMs are built on) is basically a massive machine that takes in a bunch of vectors (numbers representing words) and passes them through layer after layer, adjusting those parameters at each step. Now scale that up to a trillion knobs and massive training data… and you get GPT-4.5 or Claude Opus or whatever tomorrow’s monster model is. So when someone says “this model has more parameters” they’re not just flexing, they’re saying: This thing has more mental bandwidth, more nuance, more ways to model how language and ideas connect. But it also means: It’s more expensive to train, slower to run, and a lot harder to control. Parameters are what make AI models learn They’re not magic but they’re the closest thing we’ve got Next time you see “1.6T parameters” in a blog post, you’ll know that’s not just a number It’s the brainpower behind the magic.

Start with the boring stuff Learn what these mean: 200: OK (you passed the vibe check) 201: Created (you made something, good job) 400: Bad Request (you fucked it up) 401: Unauthorized (go touch some credentials) 403: Forbidden (you’re not allowed in) 404: Not Found (it's gone or never existed) 500: Internal Server Error (they fucked it up) it’s geek shit but you are going to see these a lot when you build.

most small businesses know social media is everything now but they still waste money on smm “agencies” who charge $2k/month to post quotes on canva meanwhile AI is cooking analyzing thumbnails, studying comments, reverse engineering virality and pumping out content 24/7 without asking for feedback you need a system and today, in the beginning of AI era you can get it built for the price of one invoice find guys like Damian in AI communities drop 2–5k get a custom agent working full-time no salary, no ego this is bitcoin 2010 energy don’t be the fool riding a dying horse while others are flying past you in automated lambos delegate as much as possible now

turns out binance has public APIs someone even built n8n community nodes for it is anyone automating crypto & trading?

log off X. touch grass. talk to real people. you’ll instantly realise how early we are in AI. folks are out here making hadron colliders in n8n while end users are still amazed the bot “remembers their name and answers stuff” it’s good to go deep but don’t overcook the steak no one needs an iPhone in the Iron Age. learn in silence. then rise like batman when the city’s ready.

is it just me or GPT-5 is losing its mind? I have multiple different chats for different purposes and it is mixing it all up saying things completely unrelated to the topic, replying with information from other chats. memory issue? is anyone else having it?

You’ve seen it a hundred times by now: DeepSeek r1 – 671B parameters GPT-4o – 1 trillion parameters Okay but what the hell are these parameters? Is this a who has a bigger d*ck competition? When people say a model has “1 trillion parameters,” they’re talking about how many tiny little knobs it has under the hood. These knobs are what the AI tweaks during training to get better at predicting text, solving problems, or just sounding smart. The more knobs it has, the more it can learn. Simple as that. Technically, these knobs are just numbers, weights and biases in a neural network. Every time the model sees a sentence, it’s adjusting those numbers. “This word should matter more, that one less. Oh, and let’s not forget this comma.” Over billions of examples, it gets good at spotting patterns, grammar, logic, tone, everything. So when it sees “I love” it knows “you” “this” are good guesses, but “pee” or “murder” might raise an eyebrow. The transformer architecture (what all modern LLMs are built on) is basically a massive machine that takes in a bunch of vectors (numbers representing words) and passes them through layer after layer, adjusting those parameters at each step. Now scale that up to a trillion knobs and massive training data… and you get GPT-4.5 or Claude Opus or whatever tomorrow’s monster model is. So when someone says “this model has more parameters” they’re not just flexing, they’re saying: This thing has more mental bandwidth, more nuance, more ways to model how language and ideas connect. But it also means: It’s more expensive to train, slower to run, and a lot harder to control. Parameters are what make AI models learn They’re not magic but they’re the closest thing we’ve got Next time you see “1.6T parameters” in a blog post, you’ll know that’s not just a number It’s the brainpower behind the magic.

“Oh wow iOS 26 🥹” while everyone’s busy licking Apple’s liquid ass, the real update just flew over their heads native automations no shortcuts. no raindance. no code. best part? you can rip out Apple’s limp "intelligence" and plug in ChatGPT like it’s a nokia battery upgrade natively you now can run automations right from your damn phone and what are most people doing? playing dress-up, changing lock screen use all brain cells hook it up to the beloved n8n automate like a psychopath airdrop it to a client like you're on Apple’s payroll who's doing this?

If you think the AI space is too crowded Here’s a goldmine nobody’s talking about: Phone automations. iOS 26 (dropping this fall) is launching full-blown Automations, basically a mini n8n built into your phone. It’s not just Shortcuts anymore. This is native, trigger based, real world automation for normies. And nobody’s ready. It will feature triggers, like n8n but much cooler: - when you connect to Wi-Fi or Bluetooth - when opening a specific app - when sending a text - when switching focus modes - when pluged into your car - when calling your mom (literally) And here’s where it gets wild You can send/receive webhooks Which means? You can chain your phone to n8n and back. Offload the heavy lifting to real backends, then return outputs straight to your pocket. This isn’t a party trick. It’s a playground for building consumer AI products that feel native, magical, and stupidly sticky. First things that come to mind: – GPT car copilots (triggered by Bluetooth in your Toyota) – productivity bots that respond to location, weather, or calendar – “focus mode” agents that coach, hype, or guide your day – instant journaling, expense tracking, meal logging, outfit planning—whatever – voice controlled workflows with no app download, just an AirDrop install $29.99/month wrappers ready to print. So what happens next? You can wait till fall, Do fucking nothing, goon over IG motivation videos Buy your new iphone, when it comes out, like a fucking mr consumer And goof around, play dress up barbie with your phone changing wallpapers OR You can learn how "Shortcuts" work at it's current state Go balls deep into automations, when it comes out during public beta in july Learn how make it "collab" with n8n Research consumer apps and learn what could be potentially selling Stack knowledge & prep plug-n-play automations This is your shot to build in a niche before it gets named. Don’t wait for launch hype. Dig your trenches now. Because this fall? You’re either cashing in or updating your lock screen.

Start with the boring stuff Learn what these mean: 200: OK (you passed the vibe check) 201: Created (you made something, good job) 400: Bad Request (you fucked it up) 401: Unauthorized (go touch some credentials) 403: Forbidden (you’re not allowed in) 404: Not Found (it's gone or never existed) 500: Internal Server Error (they fucked it up) it’s geek shit but you are going to see these a lot when you build.

I just sent a website draft to a client for review But instead of making them send emails, screenshots, or voice notes full of "change that blue thing near the what’s it called" I gave them a little chatbot. Right there on the website Built in Voiceflow. Very basic prompt Took like 20 minutes to set up. Client clicks through the site, talks to the bot as if they’re texting me “Move this button up” “Make this font bolder” “Can we try a darker green?” The bot logs everything into Airtable in real time. No back and forth No deciphering vague emails No screenshots with red circles Just clean, structured feedback Straight into my build pipeline. Saves me time Saves them effort Feels like a game Best part? Feedback feels “fun” instead of being perceived as "extra shit I need to do" Not a flashy agent Just a small, boring automation in the exact right place.

turns out binance has public APIs someone even built n8n community nodes for it is anyone automating crypto & trading?

is it just me or GPT-5 is losing its mind? I have multiple different chats for different purposes and it is mixing it all up saying things completely unrelated to the topic, replying with information from other chats. memory issue? is anyone else having it?

most small businesses know social media is everything now but they still waste money on smm “agencies” who charge $2k/month to post quotes on canva meanwhile AI is cooking analyzing thumbnails, studying comments, reverse engineering virality and pumping out content 24/7 without asking for feedback you need a system and today, in the beginning of AI era you can get it built for the price of one invoice find guys like Damian in AI communities drop 2–5k get a custom agent working full-time no salary, no ego this is bitcoin 2010 energy don’t be the fool riding a dying horse while others are flying past you in automated lambos delegate as much as possible now

as time goes on, you start seeing it clear there’s people who consume new prompt, new yt video, new AI hack watch this, scroll that, they feel busy but 0 gets done and there’s people who just build no bs, no fakery, no fluff the internet’s tricky like that flashy posts, filtered wins, curated success but here’s the thing looking successful in front of strangers will never beat actually being successful in real life especially when the people clapping don’t even know your name

log off X. touch grass. talk to real people. you’ll instantly realise how early we are in AI. folks are out here making hadron colliders in n8n while end users are still amazed the bot “remembers their name and answers stuff” it’s good to go deep but don’t overcook the steak no one needs an iPhone in the Iron Age. learn in silence. then rise like batman when the city’s ready.