#agentic-evaluation

The latest AI and developer news about #agentic-evaluation, with the original source and a short summary.

Self-Modeling Interventions Modulate Emergent Misalignment in LLMs

Research shows that interventions targeting a model's self-model, such as self-recognition fine-tuning and self-report interleaving, can modulate Emergent Misalignment (EM) in LLMs like GPT-4.1 and Qwen2.5-32B-Instruct, with prevention being harder than reversal.

·1·2026.10.06 09:00

Scaling agentic evaluation for SWE-bench

Surviving 200,000 evaluations means even agentic benchmarks need multi-tenant infrastructure.

·2026.01.08 18:00