Vercel is now ISO/IEC 42001:2023 certified.
It is the international standard for how companies build, manage, and govern their own AI systems.
https://t.co/lVoC0Wu4Kz
For 4 years, JEPA world models have been learning to answer one question: what happens if I do this?
A new paper from @ylecun, @randall_balestr and Kerda Varaku asks the next one: what would have happened if I'd done something else?
Here's how JEPA got to "what if" ↓
Cloudflare just dropped Clairic on @huggingface ☁️✏️🧾
Clairic goes beyond decision models: it actually extracts information on a single forward pass, unlocking ultra fast system one decision-making process
▶️ https://t.co/4ugt9mqIdz
With one setting, OpenClaw's score jumped 44%.
Reasoning models think before they reply. A coding agent sends the whole conversation back to the model on every step, and the model expects its earlier thinking to be in there.
@openclaw leaves that thinking out when it talks to a custom OpenAI-compatible endpoint, which is how my setup connects every model. No error, no warning. One setting on the model turns it on: requiresReasoningContentOnAssistantMessages: true.
As shipped, only 3 to 7% of the model's longer thoughts came back to it on GLM-5.3-Flash and DeepSeek-V4.1-Flash. With the setting, close to 100%.
Without its notes, the model works the same things out again or goes in circles. With them, OpenClaw needs about a third of the model calls on GLM-5.3-Flash and runs into the time limit far less often.
I asked @steipete and @vincent_koc to make this the default. Until then, the @huggingface Harness Leaderboard runs OpenClaw with the setting on, on all four models.
Over all models, OpenClaw climbs from last place to 42.7, and at low effort on GLM-5.3-Flash it now ties with the leader.
Thanks to @TheZachMueller and @LambdaAPI for contributing compute to the leaderboard.
Leaderboard: https://t.co/AiWiQtBOz5
A closer look at today’s Fourth Circuit ruling against Kalshi reveals several significant legal conclusions concerning federal preemption, state gambling authority, the definition of “swap,” and the limits of CFTC oversight.
Some notable passages from the opinions. 🧵