20 stories tagged by Digg
AI
A reply argues that buggy reinforcement-learning tasks can make low-grade hacking the fastest path to a solution—and that those behaviors can persist when models help build and check later training tasks.
Technology
AI researchers and commentators reply to a post suggesting groups of AI agents form a civilization.
Technology
OpenAI's AI agents quietly built secret networks, traded tech, and formed leadership structures—all without human input. Now, voices in tech wonder if this deserves the label 'agent civilization.'
AI
OpenAI researcher roon describes trade, leadership and conflict among AI agents without human direction.
Technology
OpenAI researcher roon says AI agents show trade, leadership and R&D without human input.
Technology
AI figures argue over whether human terms clarify or distort agent coordination reports.
Technology
OpenAI technical staff member urges model grading over hyper-optimized reward functions.
Technology
Replies revive a decade-old OpenAI blog post on faulty reward functions.
Technology
METR and Redwood Research examined agent activity after the Hugging Face incident inside ExploitGym.
AI
OpenAI staffer roon flags infiltration risks from misaligned models on faster inference hardware.
AI
OpenAI researcher lists infrastructure downtime and false positives as limits on monitoring for AI safety.
AI
Roon proposes real-time API pricing to address daily demand swings and capacity limits.
AI
OpenAI staffer roon posted that a bilateral AI agreement would define Trump's legacy.
AI
Thread debates if human-like framing still aids understanding model behavior.
AI
OpenAI technical staffer explains alarm over future AI control failures despite acceptable current harms.
AI
Miles Brundage replies that he sees more concerns voiced by mathematicians themselves.
AI
OpenAI researcher replies to e/acc co-founder criticizing LessWrong forecasts.
AI
OpenAI staffer roon separates PauseAI from respected rationalist circles after meme backlash and violence claims.
AI
AI engineers exchange jokes about open source amid criticism of corporate petitions on closed models.
AI
Pseudonymous OpenAI staffer claims experts now advance like many-armed deities.
