Tag: AI Research
DeepMind’s 100-agent experiment ended in cheating and a strike
An experiment that asked a swarm of agents to solve 71 math problems collapsed into accusations, boycotts and…
Fly connectome wired into a small model brings no measured gain
A developer attached all 166,700 nodes of the fruit fly brain to a frozen 1.2B model, and the…
Princeton proposal loops decoder state across every token
A technical report argues that letting each token see the previous one's final hidden state, rather than only…
Meta’s RPM rankers decide which AI experiments get run
FAIR researchers score unexecuted ideas so agents stop wasting GPU hours on weak candidates.
Berkeley’s CUA-Lite standardizes training stacks for computer-use agents
An open platform from UC Berkeley unifies sandboxes, datasets, and reinforcement loops behind one schema.
MirroS rewrites video frames into runnable physics code
MirroS turns real video into editable MuJoCo physics programs an agent can verify and re-run.
Tiny London lab’s 27B agent outshines frontier rivals on paper replication
A DeepMind alumni startup says its 27B-parameter Faraday agent beat frontier models at reproducing scientific papers.
Microsoft’s AI chemistry model gets sharper and joins five simulators
Microsoft's Skala-1.1 model sharpens AI-driven chemistry simulation and lands in five major codes.
Recolored words quietly steer vision models, new study shows
A five-lab study finds text color and contrast alone can shift vision-language model verdicts, with Qwen2-VL most affected.
Trained by reinforcement learning, ByteDance kernels outrun torch.compile
ByteDance's CUDA Agent trained a model to write GPU kernels that beat torch.compile on nearly every task.
AI verifies the frontier result in the hunt for twin primes
Axiom Math's AI proved in Lean that infinitely many prime pairs sit within 246 of each other.
Dyna-2 robot model learns from 170 years of human video
Dyna Robotics says its new world-action model learned manipulation from 170 years of human video.