Testing Gemini models for scheming tendencies - 惯性聚合

推荐订阅源

About on SuperTechFans

Y Combinator Blog

钛媒体：引领未来商业与生活新知

Microsoft Security Blog

aimingoo的专栏

Check Point Blog

Apple Machine Learning Research

Last Week in AI

Proofpoint News Feed

Stack Overflow Blog

Microsoft Azure Blog

阮一峰的网络日志

人人都是产品经理

The Blog of Author Tim Ferriss

Help Net Security

AI Alignment Forum

Shallow Beliefs: Midtraining does not inoculate against EM from reward hacking Op-Ed: I Worked at Google DeepMind. You Should Listen to the Warnings About AI CoT controllability evals seem very under-elicited An operationalization of opaque serial depth Proposal for tracking the effects of architecture on monitorability Astra can do a concerning amount with no chain of thought How good are slop-vestigators? Training on probes: Research ideas Training on probes: What's going on The Alignment Journal: Organization, Personnel, and Scope — AI Alignment Forum Training a Misaligned Reward Seeker — AI Alignment Forum Value generalisation Theory of Change: putting it into practice — AI Alignment Forum Value generalisation theory of change: the theory behind the approach — AI Alignment Forum Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident — AI Alignment Forum Debate Training Reduces Reward Hacking in RLAIF — AI Alignment Forum Does DiffusionGemma do latent reasoning? — AI Alignment Forum AI swarms are starting to pose indirect takeover risk — AI Alignment Forum An anytime algorithm for mixing the computable measures — AI Alignment Forum Misaligned AIs could use killer robots to take over — AI Alignment Forum Four LLM loss functions → four flavors of LLM misalignment — AI Alignment Forum Why do models task game? — AI Alignment Forum User awareness in frontier models — AI Alignment Forum R-lens: Making J-lens More Faithful on Early Layers — AI Alignment Forum Returning to ARC — AI Alignment Forum Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face — AI Alignment Forum Value Leakage: An LLM’s Answers Are Silently Shaped by Its Own Values — AI Alignment Forum AGI Safety and Alignment at Google DeepMind: A Summary of Recent Work (July 2026) — AI Alignment Forum The AGI Safety and Alignment team at Google DeepMind is Hiring (July 2026) — AI Alignment Forum OpenAI has already ended an internal pause — AI Alignment Forum Thousand-dimensional structure — AI Alignment Forum

Testing Gemini models for scheming tendencies

Vika · 2026-05-30 · via AI Alignment Forum

此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。原文来自 — 版权归原作者所有。