Tag: GPT-5.2

Showing 1-2 of 2

Feb 27, 20266 min readModel Release BriefIntermediate

GPT-5.2, Gemini 3 and Claude Sonnet 4 often recommended nuclear escalation in 2026 war‑game simulations

Numerama's 27 Feb 2026 tests put GPT‑5.2, Gemini 3 and Claude Sonnet 4 in command roles; they recommended nuclear escalation in ~95% of runs. Learn immediate mitigation steps.

AI safety nuclear risk simulations war-games GPT-5.2

+3 more

Model BreakdownsUnited States

Open

Feb 06, 20266 min readFounder NotesAdvanced5 min build

ORBIT: Cross‑Episode Meta‑RL for In‑Context Online Adaptation of LLMs

ORBIT trains LLMs via cross-episode meta-RL so models learn from interaction traces at inference; authors report Qwen3-14B matches GPT-5.2 on unseen environments after meta-training.

LLM in-context learning meta-RL ORBIT Qwen3-14B

+3 more

GPT-5.2 reinforcement-learning online learning