benchmarking agents by their ability to play video games
RESEARCH
-
U.S. Government Unprepared for World Models AI Technology
By
–
The U.S. wasn't ready for ChatGPT. Are we repeating the mistakes with world models? “Policymakers, government officials — many of them don’t even know what a world model is.” @StanfordHAI Executive Director @russellwald said on @politico
. Read more: -
AI Index 2026: AI Driving Scientific Discovery Beyond Research Tools
By
–
In #AIIndex2026's Science chapter, the report shows AI driving more scientific research, moving beyond a research tool that helps write papers or check numbers and toward actual discovery in science. Check out one of the AI Index's newest chapters here: https://t.co/62R5xYAiI5 pic.twitter.com/PZMxjOqPyW
— Stanford HAI (@StanfordHAI) 22 avril 2026In #AIIndex2026's Science chapter, the report shows AI driving more scientific research, moving beyond a research tool that helps write papers or check numbers and toward actual discovery in science. Check out one of the AI Index's newest chapters here: https://
hai.stanford.edu/ai-index/2026-
ai-index-report/science
… -
MIT Creates AI-Controlled Artificial Muscle Fibers Without Motors
By
–
MIT researchers have built AI-controlled artificial muscle fibers that mimic real human movement—no motors, just electrically driven fluid dynamics. https://t.co/zGxtJG85Va
— AI (@DeepLearn007) 22 avril 2026MIT researchers have built AI-controlled artificial muscle fibers that mimic real human movement—no motors, just electrically driven fluid dynamics.
-
New benchmark PerfectSquashBench for image model anchoring in AI
By
–
Image models tend to get much more stuck on a particular direction than text models, requiring clearing the context window fairly often. PerfectSquashBench is my new measure of how image models anchor. The squash remains merely fine after many attempts.
-
Odyssey-2 Max: Largest World Model with Real-Time Physics
By
–
🌍 Odyssey just launched Odyssey-2 Max — their largest world model yet
— Futurepedia – Learn to Leverage AI (@futurepedia_io) 22 avril 2026
It simulates physics in real time: gravity, collisions, fluids, biomechanics — all running live as you interact with it.
Scores highest in physical accuracy among all evaluated world models, with 3x the… pic.twitter.com/OiW66rgzysOdyssey just launched Odyssey-2 Max — their largest world model yet It simulates physics in real time: gravity, collisions, fluids, biomechanics — all running live as you interact with it. Scores highest in physical accuracy among all evaluated world models, with 3x the
-
Recurring AI incidents tracked over thousand occurrences
By
–
This has in fact (at various levels) happened more than 1000 times. Check out @DamienCharlotin
’s tracker and my substack essays about it. -
Agents That Self-Generate World Knowledge via Outcome-Based Rewards
By
–
How far are we from agents that can self-generate world knowledge? The work proposes an outcome-based reward that measures how much an agent's self-generated world knowledge actually improves its task success rate. The external guidance is then removed at inference. Result: A
-
NVIDIA NeMo RL Accelerates Agentic Performance with FP8
By
–
Improve agentic performance with accurate RL post-training on low-precision FP8. NVIDIA NeMo RL, an open-source library within NVIDIA NeMo, supports FP8 to speed up RL workloads by 1.48x on Qwen3-8B-Base—enabling faster iterations for agentic tool use and multi-step