Our internal data shows Claude is accelerating AI development—a possible path to recursive self-improvement, or AI autonomously building a more capable successor.
It’s happening faster than we thought, and the implications deserve greater attention. https://t.co/OVVPJO7VQx
LeWorldModel: all 4 worlds reproduced on an Apple Silicon Mac (MPS). Reacher — the last one — at 76%. Clean pattern: the 2
MuJoCo worlds lag most, since macOS forces glfw instead of egl rendering. Report + episode clips 👇
https://t.co/IJzyZ08IUT cc @randall_balestr
Reproduced LeWorldModel (JEPA + SIGReg) benchmark on Apple Silicon! 🧊
Scored 66% success on the OGBench-Cube task, proving local MuJoCo planning is viable on Mac despite the EGL/GLFW render gap.
https://t.co/CZXiAVMKxf
Gotcha for anyone reproducing: pin transformers<5. 5.x renamed the ViT attention keys (q_proj/layers.N), so stock load_pretrained rebuilds an encoder that silently mismatches the released weights → 4% instead of 90%. Fix + details: https://t.co/ICUutwJfYn
Reproduced LeWorldModel on PushT: 45/50 = 90% (eval seed 42) on Apple Silicon / MPS — in line with the paper’s 96±2.8. Planning time held up too (~2s/plan on M1-MPS vs <1s on GPU). cc @lucasmaes_@randall_balestr
Would you like to join the research effort on JEPA and World Models easily?
After a full year of hard work, we’re excited to finally release stable-worldmodel:
an open-source, scalable platform built to accelerate JEPA & World Model research!
📄: https://t.co/gnxGvens5A
Australia, Canada and the EU praised Iranians protesting peacefully and condemned killings, violence, arbitrary arrests and intimidation by authorities, urging an immediate end to lethal force by security forces and respect for free expression and assembly.
https://t.co/eelXo0ekfv
@MrChallinger@Tesla@cybertruck I’ve intervened 3 times on my X running v13. Same scenario. Definite regression. Don’t assume situations that worked great in the past will still work great in new versions. Tear each new version like a first time use.