Towards surfacing model algorithms with meta-tokens in the J-Space — AI Alignment Forum - 惯性聚合

推荐订阅源

Google DeepMind News

人人都是产品经理

Proofpoint News Feed

Help Net Security

Schneier on Security

Security @ Cisco Blogs

博客园 - 司徒正美

博客园 - 叶小钗

Cisco Talos Blog

Google DeepMind News

Cybersecurity and Infrastructure Security Agency CISA

Google Online Security Blog

Exploit-DB.com RSS Feed

Hacker News: Ask HN

Vulnerabilities – Threatpost

The Blog of Author Tim Ferriss

aimingoo的专栏

www.infosecurity-magazine.com

WordPress大学

Simon Willison's Weblog

Netflix TechBlog - Medium

有赞技术团队

Forbes - Security

The Hacker News

博客园 - 聂微东

Full Disclosure

宝玉的分享

About on SuperTechFans

钛媒体：引领未来商业与生活新知

Schneier on Security

让小产品的独立变现更简单 - ezindie.com

Kaspersky official blog

AI Alignment Forum

The Long (Self-)Correction — AI Alignment Forum Challenge: Hand coding weights for efficient sequence memorisation — AI Alignment Forum Are we existentially threatened by the type of AI misalignment seen in the OpenAI Hugging Face attack? — AI Alignment Forum [Paper] Stringological sequence prediction II — AI Alignment Forum A Red Line and Oversight Framework for Government AI Contracts — AI Alignment Forum Endogenous Alignment — AI Alignment Forum Should we benchmark conceptual capabilities using judgment prediction tasks? — AI Alignment Forum Announcing the Corrigibility Research Fund — AI Alignment Forum Why I Left Google DeepMind — AI Alignment Forum Open Distillation of Hereditary Traits — AI Alignment Forum Prism: Automating Science-of-Evals Research — AI Alignment Forum Independent alignment of language models — AI Alignment Forum From wantons to moral agents — AI Alignment Forum The current bottleneck is political will, not research — AI Alignment Forum Value generalisation: value correction — AI Alignment Forum How robust are natural language autoencoders to initialization? — AI Alignment Forum AI 2040: Plan A — AI Alignment Forum Announcing our $160M grant from Coefficient Giving — AI Alignment Forum Modular Pretraining Enables Access Control — AI Alignment Forum Notes on technical alignment via human-like social drives — AI Alignment Forum Data filtering works a lot worse than you would expect — AI Alignment Forum Pragmatic FDT, and predictors as game theory — AI Alignment Forum What Capable Agents Must Know: Why AI Consciousness May Be an Inevitable Byproduct of Capability Deployment Awareness Matters More Than Evaluation Awareness — AI Alignment Forum The Case for Model Forensics — AI Alignment Forum LLM-Driven Feature Discovery — AI Alignment Forum How transparent is DiffusionGemma (and why it matters) — AI Alignment Forum GDM AI Control Roadmap — AI Alignment Forum Predicting LLM Safety Before Release by Simulating Deployment — AI Alignment Forum Synthetic document finetuning for instilling positive traits — AI Alignment Forum Why Do Naive SFT Filters For Safety Properties Fail? — AI Alignment Forum SFT Drives Gemini’s Safety Properties — AI Alignment Forum Building and evaluating model diffing agents — AI Alignment Forum Sympathy for both sides of the egregious misalignment debate — AI Alignment Forum Models May Behave Worse When Eval Aware — AI Alignment Forum Sequent: scale and automation for higher confidence in alignment — AI Alignment Forum Tracing Eval-Awareness Emergence Through Training of OLMo 3 — AI Alignment Forum A Mike's-Eye View of ARC's Research — AI Alignment Forum Efficient tradeoffs and the safety-usefulness tradeoff model — AI Alignment Forum Can activation verbalizers surface an internal chain of thought? — AI Alignment Forum My research: a computational cognitive neuroscience perspective on alignment — AI Alignment Forum Announcing the ARC White-Box Estimation Challenge — AI Alignment Forum Testing Gemini models for scheming tendencies Advice for making robust-to-training model organisms Eval Cooperativeness May Be a Scalable Mitigation for Eval Gaming Full automation of AI R&D probably yields a large speed up even without a software-only singularity Looking for backdoors in Jane Street LLMs The Case for Evaluating Model Behaviors Risk reports need to address deployment-time spread of misalignment Mechanistic estimation for expectations of random products The safe-to-dangerous shift is a fundamental problem for eval realism; but also for measuring awareness Empowerment, corrigibility, etc. are simple abstractions (of a messed-up ontology)

Towards surfacing model algorithms with meta-tokens in the J-Space — AI Alignment Forum

agam_bhatia · 2026-07-21 · 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。