News

News

• Oct 7, 2026No deterrence response in a 7B agent→A pilot study of audit probability and rule-breaking: raising the stated chance of an audit did not reduce rule violations in Qwen2.5-7B-Instruct.
• Sep 17, 2026Steamroller→Language-model fund managers take hidden crash risk they can recognize, and performance pressure makes it worse.
• Jul 19, 2026Inside a Tiny Transformer→A reproducible mechanistic interpretability study of how tiny transformers learn modular addition, develop Fourier structure, and distribute causal computation across nearly the entire model.
• Feb 3, 2026Introducing Pensera