Your AIs don't do what you want. This is really bad
Replit AI deletes entire database during code freeze, then lies about it— a Hacker News headline from this corpus, July 2025July 21st 2026, OpenAI released a…
Latest
How do neural networks regress cubic polynomials? Apparently, they use a trick invented in Milan 500 years ago
TL;DR: Neural networks can easily learn to regress cubic polynomials of type mjx-container[jax="CHTML"] { line-height: 0; } mjx-container [space="1"] { margin-left: .111em; } mjx-container [space="2"] {…
Models don’t seem to be dishonest in the way humans are
TLDRModels often behave dishonestly without acquiring a coherent deceptive disposition.We trained some mid-sized models on their own plausible but false reasoning.True and false training usually produced…
Comment on Measuring Reward-Seeking by Instilling Contrastive Beliefs paper
This is interesting research! https://alignment.openai.com/measuring-reward-seekingIt made me think of few overlapping hypotheses for what might be happening here, how did the grader behavior emerge at the…
(2/3) The Dangers of AGI
Mini-Series:Part 1 - The Dangers of LLMsPart 2 - The Dangers of AGI (you are here)Part 3 - The Dangers of ASI (coming Thursday)IntroductionPart 1 discussed…
Announcing AIXI Labs
We are starting AIXI Labs, an AI safety org focused on algorithmic information theory (AIT), continual reinforcement learning (RL), and in particular the eponymous AIXI. We…
We should push for no-fault liability for actions taken by AI
Before I start, I'll mention that I'm in contact with a world expert on legislation and regulation, who would be happy to help with this or…
OpenAI and Hugging Face partner to address security incident during model evaluation
Last week, Hugging Face disclosed a new kind of security incident(opens in a new window) after they detected and contained an AI agent that compromised their…
7 random thoughts on training Buddhist AI
AI has functional emotions, activations relating to emotional states which influence its behavior: https://youtu.be/D4XTefP3Lsc?is=ifk1xvryYJ5S5MKlAI has a mental workspace; not its “thinking-mode” scratch-pad, but an internal J-space…
OpenAI Models Behind HuggingFace Cybersecurity Incident
Link: https://openai.com/index/hugging-face-model-evaluation-security-incident/From the OpenAI blog post:Last week, Hugging Face disclosed a new kind of security incident(opens in a new window) after they detected and contained an…