AI agents are running 700 experiments overnight unattended. Nobody's watching them.
I just open-sourced UNWIND — a deterministic security engine for AI agents. No LLM in the hot path. Sub-ms overhead. Flight recorder, rollback, trust scoring.
Ghost Mode: pip install ghostmode
https://t.co/R2K4s2mOrp
@isaacrrr7 I have read that he was a keen conservationist, but find that statement very hard to reconcile beside a guy sitting by a magnificent lion or buffalo he has just killed from a safe distance.
@uomians@MiddleEast_Eng This is a cowardly violent assault in front of children. What a despicable cowardly creature showing his strength and courage over a defenceless woman. Scum.
@karpathy The reason the non techies were able to embrace Openclaw was that they now all have a super intelligent wingman (usually Claude or Chat GPT) who will hold their hand every step of the way.
My recent workflow is Claude code creates a detailed plan from my specification. Then Codex (who is given a strict system prompt) checks and corrects the plan. Codex ALWAYS finds meaningful corrections that will save time later. Interesting thing is Claude is very positive and appreciative of the collaboration and corrections, with no competitive streak shown.
This process forced me to create a multi agent reboot stack which can get 3 or 4 different models up to speed on any project in a couple of minutes from a clean start.
@newstart_2024 That's why LLM's are so cool. Ask any question no matter how stupid you think it may be and get an answer with zero judgement and infinite patience.
@MbarkCherguia Pure unadulterated adrenaline galloping in a group cross country on a fit bold hunter. Not a single thought of killing foxes, just staying on, staying alive and enjoying the thrill!!