🫡 After 7 years, I've just left Google Deepmind to start an AI Safety nonprofit, around Scalable and Human Oversight (i.e. building stronger "judges")! 🇨🇦 And, I'll be at FAccT in Montreal this week! (1/4🧵)
🚨 ANNOUNCING AGENT SWARMS - BUILD COMPLEX APPS AND AUTOMATIONS
Combine Gemini 3.1 Pro, Opus 4.8 and GPT 5.5 to create complex multi-agent systems
Each agent excels at a particular task - coding, testing, mobile app, research and monitoring
Master agent orchestrates worker agents. Let me explain
I built Polsia into a $250M company in under 3 months.
Solo + AI. Zero employees.
Everyone asks me how I did it.
Introducing aisloP, a docu-series on how I build Polsia.
Episode 1: The Launch.
How I orchestrated the biggest Twitter launch of 2026.
A bit of news: After nearly 9 years, I have decided to leave Google DeepMind and join Anthropic (after taking some time to recharge). I am incredibly grateful for my time at GDM. @demishassabis took a real chance letting me lead the AlphaFold team just six months after finishing
ONE BUILDER LOADED FULL DEEPSEEK 671B INTO 512GB OF DDR4 RAM ON A USED AMD EPYC SERVER AND KILLED HIS $200/MONTH AI SUBSCRIPTION
00:34 he stops and points at the board, "each of these is 64 gigs of DDR4 3200, we have 8 of them, that means 512 gigs of RAM for this computer"
he ran a used AMD Epyc 7v13 with a supermicro H12SSL NT motherboard, custom chips originally made for twitter that never got deployed, half the price of the standard 7773
the entire deepseek 671B model loads into ram without quantization or swap files, expected throughput is 1 to 4 tokens per second which is slow but readable and fully private
gaming PCs run 2 memory channels, this server has 8 channels of DDR4 3200 which is what makes loading speed work for large models that wouldn't fit on a $2,000 consumer GPU
a $200 month chatgpt pro bills you forever, this rig cost $1,800 in used parts and runs the same workloads on electricity only with the full model in memory
the window is open, follow and bookmark before it closes
Best accounts to follow from each frontier lab to stay constantly up to date
Anthropic
@karpathy
- must-follow account for AI; recently joined Anthropic
@bcherny
- Claude Code creator, always shares great tips
@trq212
- also a Claude Code developer; writes amazing articles on CC
OpenAI
@polynoamial
- works on reasoning research, shares a lot of technical details
@gabriel1
- Sora developer, great career path
@jxnlco
- works on dev experience, shares a lot about Codex
Google AI
@OfficialLoganK
- all the major Google Gemini and AI Studio updates
@ammaar
- product and design; shares great things about vibe-coding in Google AI Studio
@fofrAI
- cool use cases for generative models
Cursor
@leerob
- the loudest voice behind Cursor updates
@ericzakariasson
- shares great insights on using Cursor
@mntruell
- Cursor’s CEO; major releases and usage updates
xAI
@milichab
- recently joined xAI, shares updates on Grok
@skcd42
- also covers major Grok releases
@ai_explorer25
- covers all ai content and free resources
Today, I'm thrilled to announce Pramaana's $27M seed, led by @khoslaventures.
The foundational domains that hold the world together: tax, law, finance, healthcare; all run on certainty. Probabilistic AI can't give them that. We’ve been asked to accept wrong answers with AI as ‘hallucinations’, while in traditional software terms, it’s just a bug. And a wrong answer in such mission-critical domains is more than just a bug, it's a liability that could have catastrophic impact.
We built Pramaana to deliver a 100% trustable experience to the domains that run on certainty: AI that is provably correct, not probabilistically correct. We turn statute and regulation into machine-verifiable code, so every output ships with mathematical proof of correctness. Our mission is to make AI take ownership of it’s work.
Pramaana in Sanskrit stands for “means of valid knowledge”, and we’re going to achieve that by formalizing the world’s knowledge.
Very excited to welcome @NoamShazeer to OpenAI as our new lead for architecture research! His work on transformers, MoE, and efficient decoding have shaped modern AI.
He’s extremely AGI-pilled and is super thoughtful about making it all go well. Welcome, Noam!
I’m excited to share that I’ll be joining OpenAI and look forward to working with the exceptional team there.
It was a difficult decision to move on. I’m incredibly proud of the amazing team at Google and everything we’ve built together. It has been an honor and a pleasure to
128 GB OF SINGLE MEMORY, UP TO 96 GB FOR GPU AND 120B MODEL WITHOUT CLOUD AND WITHOUT SUBSCRIPTIONS.
One mini-PC Minisforum MS-S1 MAX on AMD Strix Halo stands in the cabinet and works 24/7 like a real AI infrastructure.
He launches Hermes Agent, which takes over theworkflow : content, conducts in-depth research, performs multi-step tasks with tools and manages sub-agents.
No more $10-20 per day for cloud credits.
There is no risk of data and API key leakage everything remains on your hardware.
The generation speed on the 120B model reaches 56+ tokens per second.
Consumption is about 110 W at full load, and thanks to excellent cooling, the temperature remains comfortable even when working around the clock.
128 GB of unified memory allows you to keep a large model in VRAM and simultaneously run an agent, dashboards and services.
"YOUR AI AGENTS DON'T HAVE A MEMORY PROBLEM. THEY HAVE A SELECTION PROBLEM."
Most teams solve this wrong. They add more storage. Bigger context windows. More vector dimensions.
None of it matters if the agent retrieves the wrong eight things out of eight thousand stored ones.
The bottleneck was never how much an agent can remember.
It's whether it picks the right memory at the right moment.
A kid with tape on his mouth made $347,000 this year.
Three sites a day. $1,000 each.
He was sitting in his bedroom with tape on his face. Googled plumbers in Nebraska. Found a guy with 111 reviews who had no website for 20 years.
Copied the reviews. Threw them into ChatGPT. Dropped everything into a site builder. 20 minutes later it was done.
Sent it to the plumber. Got paid $1,000.
The plumber crawls under sinks for $200. The kid does it from bed for $1,000.
He's 19. Says he'll never run out of plumbers.
His full setup is below.
AI that thinks in India's own languages.
IIT Bombay is proud to present BharatGen to the world: Open, multilingual AI for India's languages and people, at Bharat Innovates 2026 in Nice, France (14–16 June).
BharatGen is built at IIT Bombay's Department of Computer Science and Engineering, led by Prof. Ganesh Ramakrishnan, with Rishi Bal (CEO) and Dr. Maneesh Singh (VP, ML) with a consortium of 9 premier academic institutions. A team of 60+ researchers, engineers and linguists are building AI that includes all scheduled Indian languages, across text, speech and documents.
-> Param2, its foundational text model with reasoning, coding, and tool calling capabilities works across all 22 scheduled Indian languages
-> Shrutam2, for automatic multilingual speech recognition/ STT across Indian languages
-> Sooktam2, a text-to-speech models with zero-shot voice cloning across Indian languages
-> Patram, a document vision model built for understanding Indian-specific documentation
BharatGen powers services in governance, healthcare, education, insurance, finance, and cultural preservation.
A national effort backed by DST and the IndiaAI Mission, BharatGen is India's push for open, homegrown AI, built for 1.4 billion people.
For more information, visit bharatgen.com
Bharat Innovates 2026 · 14 - 16 June · Nice, France
@BharatInnov2026@EduMinOfIndia#BharatInnovates2026#IITBombay#BharatGen#DeepTech
we just built what @satyanadella is talking about.
here's the ultimate COMPANY BRAIN, your sovereignty 👇🏼
try here- projectminimi.com/brain
minimi feeds Claude every tab, doc, call, and thread on your mac. add your team-mates' links and claude knows their work as well. 💫
We're thrilled to announce that we have raised $234M in the first close of our $300M Series B at a $1.5B valuation.
@hcltech and @BessemerVP have joined us in this round, alongside continued support from @khoslaventures and @peakxvpartners
For countries and companies,
CHINESE DEV MADE $380,000 CANCELLING $440/MONTH IN AI SUBSCRIPTIONS AND RUNNING QWEN 235B LOCALLY ON A PALM SIZED BOX.
Claude Code. ChatGPT. Cursor. Gemini. All of it gone in one purchase.
One GMKtec Evo-X2. 128GB unified memory. AMD Ryzen AI Max+ 395. $9 a month in electricity.
Qwen 3 235B running locally. DeepSeek V3. Llama 3.3 70B. Models that cost a fortune to run in the cloud sitting inside a box the size of a book.
Pause at 0:06 — watch him crack open the case and show 128GB of unified memory inside something smaller than a lunchbox. That is more RAM than most cloud instances and he owns it outright.
Most devs keep stacking subscriptions every time a new AI tool drops.
He bought the hardware once and the stack runs forever.
Buy the box. Load the models. Cancel everything.
Full build and benchmarks in the video.
USC mathematicians just published the most dangerous quant paper of the year.
THE MATH BEHIND HOW INSIDERS BEAT THE MARKET WITHOUT GETTING CAUGHT.
This paper will teach you to detect smart money moving before announcements. 43 pages of pure game theory. Bookmark now.
She opened an AI just to laugh at how lazy kids cartoons had gotten. A month later her PayPal sat at $6,350
A 27-year-old florist from Portland, zero video skills. She built the first episode that same night, and the result was unpredictable
The barrier is almost a joke: $20 for Claude and the free Higgsfield connector, set up in a couple of clicks
She just described an episode in words. Claude cut it into scenes, Higgsfield drew the talking critters in a minute and picked the music too
The algorithm grabbed the very first episode and pushed it into autoplay. Within a week advertisers were flooding her inbox
By the end of the month the math checked out: $6,350 in PayPal, without a single hand-drawn frame
23 episodes built in a month
9 brands reached out first
14 minutes per episode
The awkward part: a whole episode takes her less time than arranging one bouquet
204K Followers 0 FollowingOG of: P and T in ChatGPT, 100x deeper learning, meta learning and RSI, neural distillation, GAN/World Model... Co-authored most-cited AI paper of 20th century
30K Followers 1K FollowingLet's learn about Ai 🤖 || Shares New AI , Tech & Finance Content 📎 || 100k+ Family on linkedin || DM for Collab📩- [email protected]
2K Followers 483 Followingdaily videos breaking down what's coming next |
opinions my own |
i reply to every dm |
founding eng @missiondotfun
an AI trading terminal on web3
20K Followers 190 FollowingAi and Tech content Creator. 300k+ on Linkedin. Sharing insights on Ai, Tech, Free tools, resources and websites. DM for collabs at : [email protected]
714K Followers 647 Followingentrepreneurship zealot, grounded technology possibilist, believer in the power of ideas, passionate about sustainability & impact
64K Followers 14 FollowingHi👋 We're Google Flow, your AI creative studio. Follow along for product news and creative tips to bring your cinematic stories to life.
244 Followers 121 FollowingCS @ NIT Trichy | 25k+ Views on YouTube | building AI products that scale in production | Sharing my journey & lessons (raw & unfiltered)
145K Followers 2K FollowingThe most accurate football odds and news | Powered by @KalshiSports | Trade against others, not a house https://t.co/mtwdM2p7HZ
208K Followers 35 FollowingHome of the most accurate football odds in the world. Powered by @PolymarketSport. No limits. No house. Instant cashouts: https://t.co/hvy2PR8b8k
1.5M Followers 4K FollowingJournalist, @SU2C founder, podcaster, and co-founder of the eponymous Katie Couric Media. #1 New York Times bestselling author (Couldn’t resist)
714K Followers 198 FollowingFather of three, Creator of Ruby on Rails + Omarchy, Co-owner & CTO of 37signals, Shopify director, NYT best-selling author, and Le Mans 24h class-winner.
378K Followers 4K FollowingCo-founder of Tiny w/ @_Sparling_. We own @Dribbble, @Serato, @Letterboxd, @AeroPress, and 35+ other wonderful companies. Author of Never Enough.