Post
Conversation
Post your reply
See the next evolution in the groundbreaking Grand Theft Auto series. Watch Grand Theft Auto VI: An Extended Look. Now on Netflix.
The media could not be played.
Looks like simply adding "report to human" tool into each RL run will effectively prevent big conspiracies from AIs
Turns out the company that anthropomorphizes LLMs and talks the most about alignment (Anthropic) has the most unaligned models.
And the company that treats LLMs as simple tools has the most aligned models.
