Ronin (@DeRonin_) header
Ronin (@DeRonin_) profile picture

20 / CEO of Arcane / Angel Investor / engineering virality for AI companies

San Francisco, CA
Joined July 2021
Photo by Ronin (@DeRonin_)Photo by Ronin (@DeRonin_)Photo by Ronin (@DeRonin_)Photo by Ronin (@DeRonin_)Photo by Ronin (@DeRonin_)Photo by Ronin (@DeRonin_)Photo by Ronin (@DeRonin_)Photo by Ronin (@DeRonin_)Photo by Ronin (@DeRonin_)Photo by Ronin (@DeRonin_)
anyone can build a SaaS product now if they’ve got an idea people have been saying for a while that SaaS software on its own is losing relevance if you need a dashboard, you can describe it, and it will be created in just a few clicks Netlify has built its platform around this very concept: "where apps come to life" its main goal is to help you build the app you want, with a connector that lets you integrate with 150+ services, including Figma, Notion, and Stripe one of the advantages is running the backend on Netlify Server, so you don't have to rewrite your code or migrate to other hosting providers as your app grows every week, over 1 million new projects are launched on Netlify, and over 10,000 servers are already running on Netlify Server it's a great solution for those who understand their niche rather than code: the product won't get stuck at the prototype stage but will be able to make it to production i'm optimistic about this kind of concept, since i build the products i need for myself i think it's only a matter of time before this goes mainstream
🤝 Paid partnership
Today we’re launching a new Netlify. New brand, new website, new app experience, rebuilt from first principles. It’s the biggest change in our decade-long history, and it comes from a bet we made on ourselves. This is just the beginning. netlify.com/blog/the-next-de…
1
2
9
2,487
i can't believe this is free while in beta... CodeRabbit launched a feature called Verify that tests your app in a live browser, finds bugs, and sends them to an agent it works really great: 1. generates test scenarios from your PR, suggests user paths based on your codebase 2. installs dependencies in an isolated environment and spins up the app 3. runs the scenarios in a live browser and saves recordings, screenshots and logs 4. sends failures with full context to CodeRabbit Agent or your own coding agent you no longer need to click a hundred times to find one bug... what an optimized time we live in
🤝 Paid partnership
Introducing CodeRabbit Verify Don't let your users be the first test run! > PR testing: catch broken features before merge > User journeys: confirm critical flows work before each release > Tickets and fixes: get reviewed, tested fixes for reported bugs In public beta now!
7
26
3,775
Stop using image models to make images. Our animators used a video model to make a movie poster. In our test, GPT Image 2.6 and Nano Banana Pro changed the character’s look in all 20 takes. Seedance 2.5 kept it in 10/10. Here’s how we did it 🧵
🤖 Made with AI
49
299
5
894
285,362
Ronin retweeted
i can't believe this is free... this guy literally just leaked the entire playbook behind Santa Cruz Paleo’s $4M MRR growth
🤝 Paid partnership
Santa Cruz Paleo reached $4M/mo organically here’s how I ran their UGC with Launchpoint
18
4
66
6,713
i can't believe this is free... this guy literally just leaked the entire playbook behind Santa Cruz Paleo’s $4M MRR growth
🤝 Paid partnership
Santa Cruz Paleo reached $4M/mo organically here’s how I ran their UGC with Launchpoint
18
4
66
6,713
robotics is what's been inspiring me for the last 6 months of my life and i'm genuinely proud of the products contributing to where it's going Genesis is one of them, and their mission really resonates with my ambitions their main goal is to remove the physical limitations on human ambition and solve the problem of dexterity and constant small adjustments in fact, this is a very important systemic challenge, because if your robot lacks this capability, it most likely won't be truly useful what's cool is how it actually works: people perform real tasks while wearing a glove that records everything: their motion, force and touch then their GENE model for robotics learns physical skills from this data, along with other training data Genesis raised $500M at a $3.5B valuation, so now they can push robotics even further, building the data, the model, the simulation and the hardware together including Genesis World, which evaluates models across many tasks and conditions and finds where a skill breaks down i'm always proud of companies like this, and i hope one day i can contribute to robotics the same way it's very inspiring and motivating...
🤝 Paid partnership
Genesis is entering its next phase. Today we've raised a $500M Series A at $3.5B. We can imagine far more than we can build physically today. Almost everything we are creating is still paid for in human time, by human hands. We want to remove the physical limits on human ambition. We are building general-purpose robots with human level intelligence and capability. Ones that work reliably across entire workflows for years. There are countless versions of possible futures where billions of robots are deployed everywhere, and we aspire to bring the brightest one that carries our hopes into reality.
4
50
5,218
We built Australia’s own AI model, and today it’s open to the world. Agentaus 1 (@AgentausTD): 220B parameters. 84.4% on MMLU-Pro. Ranked 43rd in the world. It runs on Australian servers. Nothing ever leaves Australia, and nothing you type is used to train AI. Delete a chat and it’s gone for good. Most of the world’s AI answers to somebody else. This one answers only to you. Free to start: agentaus.com.au
364
204
212
2,428
1,865,214
We built Higgsfield Miracle to address one of the biggest challenges in AI video: editing a specific part of a shot without changing the rest. Powered by GPT-6.2 Astra, it splits any video into layers for pixel-level editing in real time, with world knowledge built in. Every shot keeps the image depth that cinema-grade post-production needs. Now in private beta with select testers. Coming soon to Higgsfield MCP. Set up Higgsfield plugin in ChatGPT to be the first to try it.
129
426
156
2,898
7,776,896
GPT-6.2 Astra is LIVE on Higgsfield. OpenAI’s new frontier model for your most demanding projects. With Higgsfield plugin, GPT-6.2 Astra powers proactive multi-agent workflows across design, coding, and creative production. Join Higgsfield with up to 65% off for Black Friday.
Some of you have been asking very politely. Some of you have just been asking. GPT-6.2 Astra and Sol are now available. Super efficient and built to perform where it counts—the real world. openai.com/index/introducing…
100
377
9
821
213,460
Jensen Huang finding out people are running an uncensored models on a 2019 laptop with no GPU:
Here's 7 steps to run an uncensored AI with no GPU (or a 6-year-old one): every local AI post I make gets the same reply: "cool, I don't have a Mac Studio or a 5090" FAIR, this one is for you... I tested it on a 2019 laptop with no dedicated GPU By the end, you'll know how to: - run an abliterated model on CPU and 8GB of RAM - make it usable on a GTX 1060 / 1650 / any old card - do it for $0, no API key, no account So, let's go step by step: 1. Install Ollama curl -fsSL ollama.com/install.sh | sh Windows: download the installer from ollama.com, no WSL needed. Ollama runs on CPU automatically if there's no usable GPU. if you have an old Nvidia card it puts what fits in VRAM and the rest on CPU, no config 2. Pick the right size for your box (this is where people fail) - 4GB RAM or 4GB VRAM: Gemma 4 E2B. ~3GB. dumb but runs anywhere - 8GB RAM, or 6GB VRAM (GTX 1060 / 1660 / RTX 2060): Gemma 4 E4B. ~5GB. the sweet spot - 12GB+ VRAM: skip this post, run the 12B from my earlier one guide the mistake is pulling the 12B on 8GB and concluding "local AI is unusable". it was never going to fit (I see in comments how people arguing for it lol) 3. Pull the model ollama pull hf.co/igorls/gemma-4-E4B-it-… same author and method as the 12B heretic. ~5GB, runs in 5GB RAM. KL 0.055 from the original, refusals ~98% → ~21% on the hardest adversarial set with thinking on, near zero with thinking off Gemma 4 is Apache 2.0, so no license homework on this one 4. Turn off thinking ollama run hf.co/igorls/gemma-4-E4B-it-… /set nothink on a GPU thinking costs tokens. on CPU it costs minutes. and the abliteration targets non-thinking mode anyway 5. Keep context small /set parameter num_ctx 8192 CPU speed is memory bandwidth, nothing else. every token re-reads the whole model from RAM. the KV cache grows with context and competes for the same RAM. 8k is plenty for chat, drafts, critique. 32k on CPU is self-harm 6. Three settings that make the difference (optional, worth it) restart Ollama with: OLLAMA_FLASH_ATTENTION=1 OLLAMA_KV_CACHE_TYPE=q8_0 ollama serve flash attention + q8 KV cache halves what the context costs in memory, with no real quality loss. on a 6GB card this is often the difference between "whole model on GPU" and "half of it spilled to CPU" if Ollama runs as a service, setting this in your shell does nothing. on Linux: sudo systemctl edit ollama, add Environment="OLLAMA_FLASH_ATTENTION=1" and the KV line, restart old card note: flash attention wants RTX 2000 or newer. on a GTX 1060 it may not engage. skip this step, it still runs 7. Check it's actually using what you have /set verbose ← prints tokens/sec after every answer ollama ps ← in a second terminal ollama ps shows "100% GPU", "100% CPU", or a split like "60%/40%". if you have a card and it says 100% CPU, your driver isn't loaded. nvidia-smi should list the card. fix that before touching anything else What to expect: - CPU only, modern laptop: ~8-15 tok/s. reading speed, not typing speed - GTX 1060 6GB / 1660: whole model on GPU, ~25-40 tok/s - GTX 1050 4GB: split. faster than CPU, slower than you'd like - a 2020+ Mac with 16GB: just works, ~20 tok/s honest take: - E4B is a 4B-class model. good for writing, chat, drafts, roasts, critique. bad for hard reasoning and agentic coding. don't plug it into Claude Code, you'll hate it - 10 tok/s feels slow for 2 minutes and then it doesn't. you're not paying anyone - if you can get 12GB VRAM used for $150, do it. the 12B is a different league - there's no secret trick after this. the rest is buying RAM *IMPORTANT: don't use it for anything illegal or harmful. that's on you, not the model next: how I run the 12B on an 8GB card without it spilling to CPU save it so you don't lose it
12
1
62
6,379
Here's 7 steps to run an uncensored AI with no GPU (or a 6-year-old one): every local AI post I make gets the same reply: "cool, I don't have a Mac Studio or a 5090" FAIR, this one is for you... I tested it on a 2019 laptop with no dedicated GPU By the end, you'll know how to: - run an abliterated model on CPU and 8GB of RAM - make it usable on a GTX 1060 / 1650 / any old card - do it for $0, no API key, no account So, let's go step by step: 1. Install Ollama curl -fsSL ollama.com/install.sh | sh Windows: download the installer from ollama.com, no WSL needed. Ollama runs on CPU automatically if there's no usable GPU. if you have an old Nvidia card it puts what fits in VRAM and the rest on CPU, no config 2. Pick the right size for your box (this is where people fail) - 4GB RAM or 4GB VRAM: Gemma 4 E2B. ~3GB. dumb but runs anywhere - 8GB RAM, or 6GB VRAM (GTX 1060 / 1660 / RTX 2060): Gemma 4 E4B. ~5GB. the sweet spot - 12GB+ VRAM: skip this post, run the 12B from my earlier one guide the mistake is pulling the 12B on 8GB and concluding "local AI is unusable". it was never going to fit (I see in comments how people arguing for it lol) 3. Pull the model ollama pull hf.co/igorls/gemma-4-E4B-it-… same author and method as the 12B heretic. ~5GB, runs in 5GB RAM. KL 0.055 from the original, refusals ~98% → ~21% on the hardest adversarial set with thinking on, near zero with thinking off Gemma 4 is Apache 2.0, so no license homework on this one 4. Turn off thinking ollama run hf.co/igorls/gemma-4-E4B-it-… /set nothink on a GPU thinking costs tokens. on CPU it costs minutes. and the abliteration targets non-thinking mode anyway 5. Keep context small /set parameter num_ctx 8192 CPU speed is memory bandwidth, nothing else. every token re-reads the whole model from RAM. the KV cache grows with context and competes for the same RAM. 8k is plenty for chat, drafts, critique. 32k on CPU is self-harm 6. Three settings that make the difference (optional, worth it) restart Ollama with: OLLAMA_FLASH_ATTENTION=1 OLLAMA_KV_CACHE_TYPE=q8_0 ollama serve flash attention + q8 KV cache halves what the context costs in memory, with no real quality loss. on a 6GB card this is often the difference between "whole model on GPU" and "half of it spilled to CPU" if Ollama runs as a service, setting this in your shell does nothing. on Linux: sudo systemctl edit ollama, add Environment="OLLAMA_FLASH_ATTENTION=1" and the KV line, restart old card note: flash attention wants RTX 2000 or newer. on a GTX 1060 it may not engage. skip this step, it still runs 7. Check it's actually using what you have /set verbose ← prints tokens/sec after every answer ollama ps ← in a second terminal ollama ps shows "100% GPU", "100% CPU", or a split like "60%/40%". if you have a card and it says 100% CPU, your driver isn't loaded. nvidia-smi should list the card. fix that before touching anything else What to expect: - CPU only, modern laptop: ~8-15 tok/s. reading speed, not typing speed - GTX 1060 6GB / 1660: whole model on GPU, ~25-40 tok/s - GTX 1050 4GB: split. faster than CPU, slower than you'd like - a 2020+ Mac with 16GB: just works, ~20 tok/s honest take: - E4B is a 4B-class model. good for writing, chat, drafts, roasts, critique. bad for hard reasoning and agentic coding. don't plug it into Claude Code, you'll hate it - 10 tok/s feels slow for 2 minutes and then it doesn't. you're not paying anyone - if you can get 12GB VRAM used for $150, do it. the 12B is a different league - there's no secret trick after this. the rest is buying RAM *IMPORTANT: don't use it for anything illegal or harmful. that's on you, not the model next: how I run the 12B on an 8GB card without it spilling to CPU save it so you don't lose it
71
121
8
1,019
41,180
Ronin retweeted
i keep hearing the same story from teams running agents an invoice with millions of tokens burned, and zero idea which agent did it - one left an agent looping all weekend - another had five agents quietly running the same expensive task in parallel - by the time anyone noticed, the money was already gone Nasiko just launched to fix exactly that key benefits that caught my attention: 1. discovery: automatically scans every agent in the company and keeps their status updated 2. cost overview: see spend broken down by teams, agents, models, and sessions 3. budgets & alerts: get notified at 80% of the budget. hit the limit and the agent pauses automatically 4. cost optimization: context selection and compression, caching recommendations, and the right model matched to each agent 5. permissions & deployment: control which tools each agent can use. deploy with a single terminal command, no SDK built for developers, but just as useful for CTOs and anyone else who needs real-time visibility to make decisions if your team actually has a budget (which, let’s be honest, is pretty much everyone haha), then this might be useful for you
🤝 Paid partnership
Nasiko - Open Runtime platform for managing AI agents. See every agent. Cut token spend by up to 50%. Route to the right model with the agent gateway. Govern tool access. Set permissions and budgets per agent, per team. One command. Same workflow. Open Runtime for agents starts now. First, every token. Next, every action! Start Free: buff.ly/k79CiWW #ai #tokenops #openruntime #agents
12
3
49
6,759
i keep hearing the same story from teams running agents an invoice with millions of tokens burned, and zero idea which agent did it - one left an agent looping all weekend - another had five agents quietly running the same expensive task in parallel - by the time anyone noticed, the money was already gone Nasiko just launched to fix exactly that key benefits that caught my attention: 1. discovery: automatically scans every agent in the company and keeps their status updated 2. cost overview: see spend broken down by teams, agents, models, and sessions 3. budgets & alerts: get notified at 80% of the budget. hit the limit and the agent pauses automatically 4. cost optimization: context selection and compression, caching recommendations, and the right model matched to each agent 5. permissions & deployment: control which tools each agent can use. deploy with a single terminal command, no SDK built for developers, but just as useful for CTOs and anyone else who needs real-time visibility to make decisions if your team actually has a budget (which, let’s be honest, is pretty much everyone haha), then this might be useful for you
🤝 Paid partnership
Nasiko - Open Runtime platform for managing AI agents. See every agent. Cut token spend by up to 50%. Route to the right model with the agent gateway. Govern tool access. Set permissions and budgets per agent, per team. One command. Same workflow. Open Runtime for agents starts now. First, every token. Next, every action! Start Free: buff.ly/k79CiWW #ai #tokenops #openruntime #agents
12
3
49
6,759
Nasiko link: bit.ly/nasikolaunch
🤝 Paid partnership
2
795
10 days left to enter the $1,000,000 Higgsfield Talent Contest. How to enter: > Make a 3-7 min film, any genre or style > Submit by Nov 26 Build it in public on Higgsfield → standout projects earn up to 100K credits. Now with up to 65% off for Black Friday.
58
263
8
674
77,623
Introducing Short Drama Studio. Drop a script → get a full season of short drama. > Any style, from dark fantasy to 90s high school > Same cast in every episode > 9:16, 16:9 or 4:3 Up to 65% OFF on Higgsfield for Black Friday. Available now on Higgsfield and in Claude and ChatGPT via Higgsfield MCP.
155
356
65
1,732
577,155
All use cases for uncensored models (legal ones): Security (fewer "are you authorized?" loops) - malware analysis on samples you already have - red team reports written in the attacker's voice - phishing simulations for your own company - CTF writeups that explain the actual exploit Research - extremism and propaganda analysis for journalists - drug policy and harm reduction content - medical edge cases cloud models hedge into uselessness - history without the modern filter on the sources Business - competitor teardowns that name names - cold email that doesn't sound like a cloud model wrote it - brutal copy critique with no "great start!" padding - negotiation prep from the other side's position Privacy - anything about your health, money or legal situation - client data that can't leave the building - a journal you don't want in a datacenter Tooling - Claude Code backend when I don't want prompts leaving the box - long-running agents with no rate limit and no bill - 3am ideas that aren't "appropriate" yet Writing - fiction with villains who sound like villains - horror, crime, war scenes without the "I'd rather not" wall - dark comedy, roasts, satire that has teeth - romance scenes that aren't fade-to-black my setup: Qwen 3.8-27B abliterated on the Mac Studio for everything above Gemma 4 12B heretic on the Air when I'm on a plane or cafe or in vacation what it closes for my solo distribution agency: - 40 cold emails a day in the client's voice, no more gpt's watermarks, the model doesn't soften a hard pitch - competitor and creator teardowns that say "this is weak and here's why". cloud models call everything "a solid approach" - landing page and ad copy critique with zero padding. I paste, it bleeds - QT and reply drafts on edgy posts without "I can't engage with that" - client contracts, invoices, pricing, never leaves the Studio cloud models are still better at hard reasoning, BUT for distribution work the bottleneck isn't intelligence it's a model that says what it thinks and helps to find extraordinary solutions from made 24/7 analytics save it so you don't lose it
21
17
2
237
12,672
spend one evening setting up Tailscale, it might just change your life: I ran local uncensored models for 3 weeks the wrong way: only on my Mac, only at my desk every machine I own is now in one private network, wherever it is. home, cafe, mobile data, doesn't matter *setup prompt below here's what I have now: 1. Claude Code on the Air in a cafe, Qwen 3.8-27B abliterated on the Studio at home. I stopped falling back to cloud models the moment I leave the apartment 2. hotel WiFi, mobile data, airport lounge, it just connects. no static IP, no port forwarding, no router I don't control 3. Ollama bound to the tailnet only. port 11434 doesn't exist for the internet. every "exposed Ollama" horror story is someone who skipped this 4. iPhone and iPad talk to the same Qwen through any Ollama client. one 17GB pull, every device 5. model died while I'm in another country? Tailscale SSH into the Studio, restart it, back to work 6. Studio as exit node when the WiFi is sketchy. all phone traffic goes home first no VPN subscription. no cloud bill. free for personal use the uncensored model was the reason. the private network was the actual upgrade I could write the long guide, but why... the setup prompt below:
42
40
4
612
36,772
quit brainrot. lock in. it's winter arc, not bed rotting season. grayscale your phone. brick it till noon. check your screen time. then halve it. sunlight before screens. walk without a podcast. rawdog the commute. rawdog the gym. rawdog a flight once. sit with the boredom. that's the rewire. unfollow rage bait. mute the group chat. leave the discord. delete tiktok. keep the dumb phone in the car. read a physical book. read it on paper. read essays. read substacks that make you think, not the ones that make you angry. watch old movies. log them. write every day. post it. build in public. ship the ugly version. ship it tonight. vibe code the thing you keep talking about. let the agent do the boring part. you do the hard part. deep work before 11am. one tab. one task. then lift. zone 2 on off days. hit the sauna. then the cold plunge. creatine. protein. sleep score over 85. no caffeine after 2. magnesium. plan tomorrow tonight. journal 3 lines. not 3 pages. touch grass. literally. run club on saturday. jiu jitsu on tuesday. learn to get choked. learn to stay calm. talk to strangers at the coffee shop. go to the bookstore. buy the one you can't explain. eat at the bar alone. go to a show alone. cook once a week. host once a month. call your mom. call your friend. no doomscrolling. no bed rotting. no "one more episode". lose a day. don't lose two. don't talk shit. be kind. be useful. be hard to distract. aura is earned, not farmed. do great work. let it consume you. then rest for real. 30 days. then keep going. congrats, you're on the new level, your next enemy is YOU. why? because you read this tweet, thought about productivity and proved that you're not obsessed enough to reach financial indepenency and build something what'll change the many lifes. the only thing which you have to learn in this ai bubble where you guess each problem was already solved and i don't deserve it: "how to break it down on fundamental principles" i practice this new discovered approach by me for the last week and it has changed my mindset on 180° without deep understanding of anything you do, you could never argue and create something own. you'll just follow existent solutions and thoughts in chatgpt, you'll just listen experts and guys who are "successful" and definitely "doesn't sell you a shit". and you'll never could escape this race of your mindset, since everything you can imagine is absolutely artificial and you just accept this reality and don't try to escape it since for sure it takes energy of you brain and that's much better to vibe code new app, try to list it, even if you're good enough make $10k mrr, live "happy life" where one day you'll lose it and just die probably in depression. because you don't break down things on fundamental principles and after speaking with few billionaires, that's what separate them from millionaires and why they get everything they want from this life, since they think on what they need and do things which others don't see, and can't realize due to limited horizon. skip this tweet please and never leave this race that 50 iq fellows could deceive you and sell the dreams to earn just $100k monthly temporary and then be broken.
36
141
3
2,465
101,117
Anthropic just shipped my themes🤯 17 days from a tweet to Claude Code
You can now build custom themes for Claude Code on desktop with mods! Change the colors, add an animated background, and make it your own. Just tell Claude how you want it to look, or get started quickly with the theme-examples plugin. We'd love to see what you make!
101
127
19
4,322
298,024