Dries Smit

@MindsAI_Jack @tufalabs Number 1 already!🔥 Well done!

@DriesSmit1 @tufalabs Yeah. At least V2. If you look at the leaderboard now, you'll see me there at the top for now. 🚀

Good luck everyone! Especially my friends at @tufalabs ! Looking forward to seeing the innovation and creativity that happens as a result of this contest.

@MindsAI_Jack Thanks!😂

Go @DriesSmit1 !

@vr4300 @symbolica Fantastic result! Is this a single run across all the games, and how long did it take to complete them? Also, was the same harness used for every game without any game-specific hints?

Unbelievable result from the @symbolica research team! In just 1 day since the launch of ARC-AGI-3 we have been able to raise the score from 0% to 36%, and within a very close margin to human baseline as well. See our blog post linked in this thread for a detailed score breakdown and replays.

1987: AI can't win at chess—planning is uniquely human 1997: AI can't win at Go—intuition is uniquely human 2016: AI can't win at poker—bluffing is uniquely human 2023: AI can't get IMO gold—reasoning is uniquely human 2026: AI can't make wise decisions—judgment is uniquely human

ARC-AGI Discord weekly Sunday community discussion (1pm ET / 18:00 UTC): chill deep-dive on 2025 results! @arcprize Friendly vibes, all welcome 🤝 discord.gg/9b77dPAmcA #ARCAGI Congrats to all the 2025 winners! Check the Hall of Fame for more fan art! 😁🎨

Congrats to Jack Cole and the team on achieving third place in ARC-AGI-2!🥳 Great work from @MindsAI_Jack for driving this effort. Thanks to @arcprize for organising another exciting competition this year. We're looking forward to the ARC-AGI-3 competition next year at @tufalabs.

We get transfered SO MANY STUFF out of the box from our ancestors, its ridiculous. My favorite example is that empirically, we seem to have "snake detection module" hard-wired to our brain (that detects snake faster than other things) We are, in so many ways, literally pretrained models

Eliezer Yudkowsky and Nate Soares wrote a book about superintelligence, called If Anyone Builds It, Everyone Dies. It was released last week. The book is now a New York Times bestseller. It is endorsed by leading scientists across many fields, including a "godfather of AI" and the most cited living scientist; by national security experts and former government officials; by journalists; by many other public figures. Max Tegmark, the MIT physics professor, calls it the most important book of the decade. Stephen Fry says it's the most important book he's read for years. No surprise. In the coming years, humanity could develop a mind smarter than all of us combined. The book makes a careful argument for why, with anything remotely like the current state of this field, no one would have any control over what this mind would be or what goals it would have, and humanity will end if it creates a superintelligence. Humanity faces a clear danger of literal extinction caused by AI. This is not an exaggeration; literal deaths of everyone and the end of basically everything of value to us in the universe. And AI companies are in a race to kill everyone. This book is so important because it could help stop humanity on this path to self-annihilation. The book of the day in The Guardian. As The Spectator puts it: "If more and more people understand the danger, wake up and decide to end the “suicide race,” our fate is still in our own hands". Now, a New York Times bestseller. We need to raise the alarm; make the careful case; make it easier for people to be brave to speak up about the danger. While humanity lives, it is still possible for us to stop. To convince the governments to make sure that no one, anywhere, can create a superintelligence until humanity knows how to do that safely. This book can serve as a wake up call for our species. I really hope this is just the beginning.

"1m context window" models after 200k tokens

@ClementBonnet16 @MindsAI_Jack @arcprize Thanks, Clem!

@MindsAI_Jack @DriesSmit1 @arcprize Congrats, Dries!!

Great writeup by Dr. Dries Smit about his winning approach to the ARC 3 Preview competition. @DriesSmit1 @arcprize @dries.epos/1st-place-in-the-arc-agi-3-agent-preview-competition-49263f6287db" target="_blank" rel="nofollow noopener">medium.com/@dries.epos/1s…

Thanks Jack! For anyone who is interested, feel free to give it a read.

I’m thrilled to share that we won 1st place in the ARC-AGI 3 Agent Preview competition! 🏆🎉 Huge thanks to @ARCPrize and @gregkamradt for organising such an exciting and tense contest. Looking forward to seeing everyone on the Kaggle leaderboard in 2026🚀 @MindsAI_Jack @tufalabs

Yesterday Ethereum turned 10. Today, lean Ethereum is unveiled as a vision—and personal mission—for the next 10 years. We stand at the dawn of a new era. Millions of TPS. Quantum adversaries. How does Ethereum marry extreme performance with uncompromising security and decentralization? TLDR: next-generation cryptography is central to winning both offense and defense. Disclaimer: This is a Drake take™ aimed at a broad audience. A technical deep dive into hash-based post-quantum signatures and SNARKs will follow. A healthy diversity of views across Protocol, the EF, and the broader Ethereum community is expected and welcome. It strengthens us. defense—fort mode Ethereum is special. 100% uptime since genesis. Unrivaled client diversity. $130B in economic security (35.7M ETH staked × $3.7K)—maybe soon $1T. Ethereum is poised to become the bedrock of the internet of value, securing hundreds of trillions over decades, even centuries. Ethereum must survive anything: nation states, quantum computers. Whatever comes. Call it fort mode. If the internet is up, Ethereum is up. If the world is online, the world is onchain. offense—beast mode Ethereum is hungry. “Scale L1, scale blobs” is a strategic urgency inside the EF’s Protocol cluster. Expect low-hanging performance gains over the next 6–12 months. Longer term? Think gigagas L1, teragas L2. Call it beast mode. → 1 gigagas/sec on L1: 10K TPS, ambitious vertical scale → 1 teragas/sec on L2: 1M TPS, sprawling horizontal scale Scale vs decentralization? Why not both. The moon math we need is now tamed: → real-time zkVMs for lean execution → data availability sampling (DAS) for lean data A delicious cherry on top: full chain verification across every browser, wallet, phone. lean upgrades Lean Ethereum proposes bold upgrades across all three L1 sublayers: → lean consensus is beacon chain 2.0: hardened for ultimate security and decentralization, plus finality in seconds; formerly branded as “beam chain” → lean data is blobs 2.0: post-quantum blobs, plus granular blob sizing for a calldata-like developer experience → lean execution is EVM 2.0: a minimal, SNARK-friendly instruction set (possibly RISC-V; pronounced “risk five”), boosting performance while preserving EVM compatibility and its network effects The consensus layer (CL), data layer (DL), execution layer (EL) have each been reimagined from first principles. Together, they unlock fort mode and beast mode. The goal: performance abundance under the constraint of non-negotiable continuity, maximum hardness, and refreshing simplicity. lean cryptography Hash-based cryptography is emerging as the ideal foundation for lean Ethereum. It offers a compelling, unified answer to two megatrends reshaping the ecosystem: → the explosive rise of SNARKs → the looming quantum threat Imagine the leanest cryptographic brick—the hash function—singlehandedly powering L1: → CL: hash-based aggregate signatures upgrade BLS signatures → DL: hash-based DAS commitments upgrade KZG commitments → EL: hash-based real-time zkVMs upgrade EVM re-execution A cryptographic jewel in each of lean CL, lean DL, lean EL. lean craft Lean Ethereum is more than a blueprint for hardening and scaling Ethereum. More than just doubling down on security, decentralisation, and cutting-edge cryptography. It is an aesthetic. An art form. A craft. Think Jiro in Dreams of Sushi. When we can go the extra mile, we do. Minimalism. Modularity. Encapsulated complexity. Formal verification. Provable security. Provable optimality. These are subtle yet important technical considerations. Stay tuned for the post on post-quantum cryptography that will make them explicit. lean legacy After 10 fantastic years, lean Ethereum is a generational oath. To keep Ethereum online no matter what. To scale it without compromise. To make it worthy of those who come next. This is about legacy. We are builders, we are missionaries. We are Ethereum. I hope you join us.

The path to solving hunger, disease and poverty is AI and robotics

@robertwiblin Mentioned in this incredible video from @robertskmiles youtu.be/JD_iA7imAPs?si…

the goal of life is to be excited to go to work and excited to go home

Should you xeet about +0.14 on the ARC-AGI-2 leaderboard? Probably not, but it's a hard benchmark! Maybe more significant gains soon. We are cooking MindsAI @tufalabs. Try it with your best new ideas. @arcprize. @DriesSmit1 @tufaben @bayesilicon @MohamedOsmanML

Jeff Clune says early OpenAI felt like spotting alien intelligence on approach. “It's like you're an astronomer.. and a handful of people have the expertise to say, aliens are on the way.” He knew they were on their way. But everything around him still looked normal. What happens when the Trisolarans arrive?