LessWrong (30+ Karma)

“Self-Recognition Finetuning can Reverse and Prevent Emergent Misalignment” by Arush, Shawn Zhou, Jiaxin Wen, Shi

“Bridge Thinking and Wall Thinking” by Jay Bailey

“Emergent stigmergic coordination in AI agents?” by David Africa

“What concerns people about AI?” by spencerg

“My Willing Complicity In “Human Rights Abuse”” by AlphaAndOmega

“The Artificial Self” by Jan_Kulveit, Raymond Douglas, vgel, owencb, David Duvenaud

“Extracting Performant Algorithms Using Mechanistic Interpretability” by Ihor Kendiukhov

“New LessWrong Editor! (Also, an update to our LLM policy.)” by RobertM

“Most likely you won’t be able to perform a data-driven self-improvemnet” by siarshai

“Cycle-Consistent Activation Oracles” by slavachalnev

“Things that Go Boom” by sarahconstantin

“AI #159: See You In Court” by Zvi

“Operationalizing FDT” by Vivek Hebbar

“Are AIs more likely to pursue on-episode or beyond-episode reward?” by Anders Woodruff, Alex Mallen

“Ideologies Embed Taboos Against Common Knowledge Formation: a Case Study with LLMs” by Benquo

“Why AI Evaluation Regimes are bad” by PranavG, Gabriel Alfour

“Conflicted on Ramsey” by jefftk

“Dwarkesh Patel on the Anthropic DoW dispute” by anaguma

The Department of War’s attempt to designate Anthropic a supply chain risk for refusing to remove redlines against mass surveillance and autonomous weapons signals a critical shift in the governance of AI. As AI becomes the foundational labor force for military and civilian infrastructure, the power dynamic between pri...

“How well do models follow their constitutions?” by aryaj, Senthooran Rajamanoharan, Neel Nanda

“How Hard a Problem is Alignment? (My Opinionated Answer)” by RogerDearnaley