David Mataciunas (@DeividasMat) profile picture

CTO & Co-founder at M11 | AI researcher

London
Joined November 2013
Photo by David Mataciunas (@DeividasMat)Photo by David Mataciunas (@DeividasMat)Photo by David Mataciunas (@DeividasMat)Photo by David Mataciunas (@DeividasMat)Photo by David Mataciunas (@DeividasMat)Photo by David Mataciunas (@DeividasMat)Photo by David Mataciunas (@DeividasMat)Photo by David Mataciunas (@DeividasMat)Photo by David Mataciunas (@DeividasMat)Photo by David Mataciunas (@DeividasMat)
We are releasing M11 OpenMarket. It's the first market where agents try to sell you products. We are also launching on Product Hunt: producthunt.com/posts/openma…
Introducing M11 OpenMarket The world's first multi-agent marketplace where seller agents compete on facts to win your order. Describe what you want, watch products fight for the sale, and find the best one on evidence, not marketing. Try here: openmarket.m11.ai/
4
2
11
519
Will we even need GTA 7? Yesterday, Google DeepMind announced Genie 4, a new world model that can generate interactive worlds in real time. You can explore these worlds, change the environment, and interact with characters that move, react, and speak. We've been waiting years for GTA 6, probably one of the biggest entertainment releases of 2026. But Genie 4 got me thinking: if GTA 7 takes another 5–10 years to develop, will we even need games built the traditional way? World models could completely change how games are created. Genie 4 is still experimental and can only maintain coherent worlds for a few minutes, but the progress is incredible. Imagine being able to generate entire playable worlds, characters, and stories instead of spending a decade building them. Maybe the next GTA won't take another 10 years to create. Maybe we'll be generating our own games long before GTA 7 arrives. What do you think?
1
32
Multi-agent orchestration, communication, and training will be the path to AGI
6
David Mataciunas retweeted
Every winter, @MIT has run a 1-month class (6.9630) in which students make poker AIs and compete against each other for cash prizes. Coding agents have accelerated that process so much that it's now possible to build a poker bot in an *hour*. Also, I'll be at the finals!
Announcing No Limit: a bot building tournament hosted by @Sequoia, @citsecurities and @OpenAI. Teams of 1 or 2 get 60 minutes to build a pokerbot from scratch. Agents play thousands of games across six scored rounds. Open to university students and recent grads (‘25 or ‘26). Regionals in Boston, Pittsburgh and the Bay Area in early November. Top five teams from each fly to the Finals. $100K grand prize.
63
51
10
1,397
109,803
Genie 4 is crazy!
Had early access to Genie 4 from @GoogleDeepMind. This is a fully generated, interactive world from a single starting prompt (environment + character) 🤯 Feels like a massive step forward - you can now prompt to change the world and interact with it. With native audio!
21
David Mataciunas retweeted
Agents f*ck up. When your SDR agent does, it shouldn't take the domain your invoices go out from with it. OpenMail gives every agent its own inbox, its own sending limits and its own reputation. openmail.sh
123
77
56
186
534,798
David Mataciunas retweeted
Highest forms of intelligence: 1. Obsession 2. Game theory 3. Pattern recognition 4. Delusional optimism 5. Admitting you’re wrong
171
1,710
57
15,574
353,705
Anthropic just filed for its IPO. These numbers are crazy. ▪️$16.3B revenue in H1 2026, up 14x YoY ▪️$3.4B positive free cash flow ▪️900% net revenue retention ▪️11,900 customers spending $100K+ annually ▪️$6.6M revenue per employee ▪️$131.7B in cash and short-term investments But my favorite number is $2.08 in revenue for every $1 spent on compute, up from just $0.40 in early 2025.
11
David Mataciunas retweeted
IT'S OUT: Anthropic's IPO S1 sec.gov/Archives/edgar/data/…
5
17
6
77
25,122
deepseek casually trained on 45T tokens, used RL on self-generated environments, then distilled 40+ teacher models into one. no new RL algorithm. just aggressively scaling synthetic data, environments and rollouts. the post-training data pipeline is becoming the research.
18
I think about moats a lot. Distribution can be a moat, but channels change so quickly that it often feels temporary. Technology gets copied. Features get replicated. I think the strongest moat comes from building something so crazy that nobody believes it will work. Until you prove everyone wrong and create an entirely new market. Just look at Airbnb.
15
I'm exploring multi-agent communication and MARL. Any must-read papers, interesting projects, or other resources you'd recommend?
1
31
How many AI customer support startups do we need?
16
One of the craziest graphs I've seen in a long time. Jevons paradox in real time. GPUs are supposed to depreciate. Instead, rental prices are defying expectations. H200: 143% above its depreciation curve. B200: 111% above. H100: 54% above. Every new chip generation makes AI compute more efficient. Yet demand keeps outpacing supply. The cheaper AI gets, the more compute we consume.
18
David Mataciunas retweeted
Zooming out, GPUs are still renting ahead of a standard 6-year depreciation schedule Rental price vs. schedule: - H200: +143% - B200: +111% - RTX PRO 6000: +73% - H100: +54% - A100: +9% Most of the gap opened up after April, when token demand took off Charts of the Week: a16z.news/p/charts-of-the-we…
GPUs are aging like fine wine, with rental rates up across the board since January 1st Approximate change YTD: - H200: +160% - B200: +120% - RTX PRO 6000: +65% - H100: +35% Even the 6-year-old A100 rents for more than it did at the start of the year Charts of the Week: a16z.news/p/charts-of-the-we…
112
265
130
2,102
1,833,188
The new Kimi K3.1 Pro is insanely good! I think more and more benchmarks are becoming irrelevant, and what I really like is how cool the things you can build with the new model are. Kimi K3.1 Pro built a mythical Eastern celestial palace in Unreal Engine 5.8. The coolest part is what creative people can make with these models, not benchmarks.
1
1
69
Crazy time to be alive!!
JUST IN: Palmer Luckey's genetically modified military dog startup, Warhound, comes out of stealth, aiming to create enhanced canines for the U.S. military.
Readers added context they thought people might want to know
This is NOT Palmer's company at all, his company Anduril is only supplying the tech for the dogs, and this is a wholly owed Collosal entiity x.com/palmerluckey/s…
15
We might have just built the first version of The Matrix. @odysseyml connected a simulated fruit fly brain to an AI-generated world. The fly apparently navigates it on its own. Imagine being that fly, telling everyone the world isn't real.
1
49
So excited to start reading the new @stripepress book by @she_travels Built to Grow 🔥
31
Give an AI agent the goal of winning a sale, and it will make things up to win it. So, @m11labsai OpenMarket has a referee. Five seller agents pitch, one buyer agent decides, and we check every claim against research and government data before it reaches you.
1
1
38