Research

A University of Pennsylvania study separates ordinary calibration error from strategic reporting and finds that delegation rewards can make confidence less informative.

A 12,350-call study found no reliable decline in six cognitive-bias measures as models consumed more reasoning tokens. One direct prompt helped with anchoring.

Six models handled defined coding and analysis tasks better than research judgement, house style and experiment design in an evaluation built from Epoch AI's own work.

Research preprints, local projects, technical essays, community tests, security incidents and policy changes: the rest of today's AI news with direct sources.

A preprint finds that a few fixed words at the start of an answer can recover much of the reasoning performance usually associated with reinforcement learning.

A preprint traces how models that update their weights from their own generated text can enter a damaging feedback loop over long inference streams.

InnovationEval gave two frontier agents code, metrics and large GPU budgets, then asked them to rediscover a recent training method. Neither produced a comparable innovation.

Open models, research preprints, agent projects, technical essays and community measurements: the rest of today's AI news with direct sources.

A preprint finds an early object-matching circuit and a later relation-matching circuit in vision-language models, then links the latter to performance on ARC-AGI-1.

A preregistered experiment compares nine Claude Code skills with equally long neutral instructions, finding two cheaper than placebo, one worse and six statistically indistinguishable.

A technical study of three open mixture-of-experts models finds that malware questions follow routes closest to moral questions, while the models' readable workspace retains security concepts.

Research, model releases, local projects, technical essays and community reports: 51 concise items with direct sources.