/explore
|
You are here 
|
www.lesswrong.com | ||
| | | | |
www.greaterwrong.com
In a recent tweet, Anthropic seems to have asserted that hyperstitionis responsible for observed misalignment in their AIs. Strangely, the research post they use as evidence appears to be only vaguely related to hyperstition[1]? I think this is part of a pattern by Anthropic of promoting the theory of hyperstition- the idea that writing about misaligned AI helps bring misaligned AI into existence- without explicitly calling it that. However, the research post shared with this tweet doesn't seem to be about writing misaligned AI into existence. Instead they find that training the model on reasoning traces- generated by reflecting on its constitution while giving users ethical advice on difficult dilemmas- reduces misaligned behavior. This presumably works by ... |
|
| | | | | ||
| | | | |
joecarlsmith.substack.com
What can we learn from recent empirical demonstrations of scheming in frontier models? |
|
| | | | | ||
| | | | |
www.darioamodei.com
Confronting and Overcoming the Risks of Powerful AI |
|
| | | | | ||
| | | | |
techrights.org
[AI summary] The text is a compilation of various topics and discussions, including commentary on internet and technology trends, software development, personal reflections, and specific events such as the 'xz' backdoor issue and the legal dispute between filmmakers and Reddit. It also mentions the use of Gemini as an alternative to the World Wide Web, along with personal and technical musings. |
|
| | | |||
