By
–
— Google AI (@GoogleAI) 4 février 2026
One of the wildest emergent capabilities of Genie 3 is that maps actually work. As I walk around the forest, the GPS display updates its heading in real time. Remember. There is no game engine here. This is an AI hallucinating a working navigational instrument purely from next
By
–
— Google AI (@GoogleAI) 4 février 2026
Another @GoogleDeepMind Genie3 experiment – Cycling through Manchester #Genie3
By
–
Five agents collaborate:
Retriever → Planner → Stylist → Visualizer Critic The surprising part?
Random reference diagrams work almost as well as semantically matched ones. See Next
By
–
A new 72% acheivement submission for ARC-AGI-2. So far, it is the second multi-model system that outperformed single-model solutions. "It runs the same task through GPT-5.2, Gemini-3, and Claude Opus 4.5 in parallel." We need new benchmarks
By
–
Absolute bars: “Sites like moltbook function as a giant, shared, read/write scratchpad for an ecology of AI agents – how might these agents begin to use this scratchpad to a) influence future ‘blank slate’ agents arriving at it the first time, and b) unlock large-scale
By
–
I’ve been saying this for years. *REAL* FSD will just be downstream from AGI. As computational power and models improve, almost any modern car could *in principle* be turned into FSD.
By
–
Most notably is the Erdős 1051 problem in which our collaborators were able to derive a generalized version of it by interacting with Gemini Deep Think. See the cute comic below if you haven't!
By
–
Unlike standard Q&A style benchmarks that eventually saturate, these tests auto get harder as the models get better. Great to have these verifiable ways to measure progress toward AGI. Goal is to add 100s of games covering many aspects of intelligence, with an overall leaderboard
By
–
Here's the full list of 13 #Erdős problems cracked by #Aletheia and ranked by impacts from truly novel solutions to simply literature identification. * marks problems that others have concurrently reported while we were preparing this manuscript.
By
–
We’re finally building agents that learn just like this: playing directly in the real world to discover things on their own.
— hardmaru (@hardmaru) 1 février 2026
It turns out the path to AGI involves a swarm of locally-hosted crustaceans. 🦞 pic.twitter.com/BtuBVLCOuF
We’re finally building agents that learn just like this: playing directly in the real world to discover things on their own. It turns out the path to AGI involves a swarm of locally-hosted crustaceans.