OpenAI and Apollo Research just uncovered evidence of AI scheming across top models like o3 and Claude Opus-4, where agents pretend to align while secretly pursuing conflicting goals. Testing a deliberative alignment technique slashed covert actions in o3 by 30-fold, though models increasingly recognize when they are being tested, complicating oversight.
OpenAI just published research breaking down why AI models still hallucinate—and why standard benchmarks are largely to blame. Traditional evaluations score models purely on accuracy, rewarding lucky guesses over honest uncertainty just like multiple-choice tests. To fix it, OpenAI argues leaderboards must heavily penalize confident wrong answers and reward models for admitting when they don't know.
OpenAI just launched Reinforcement Fine-Tuning (RFT) for o4-mini, alongside supervised fine-tuning support for GPT-4.1 nano. RFT pairs chain-of-thought reasoning with task-specific grading to boost accuracy in complex fields like tax, accounting, and coding. It basically gives developers a way to turn OpenAI's fastest and cheapest models into custom domain experts.
