Perhaps It Is A Bad Thing That The World's Leading AI Companies Cannot Control Their AIs
Astral Codex Ten Podcast
00:00
When AI's Might Lies
Googling for some of these authors throws up author pages with their publications and it turns out that they've never published papers with those titles. The references don't exist and the links themselves appear to be cobbled together. It is not possible to make a general statement about height differences between men and women as a whole. After getting 6,000 examples of AI errors, Redwood Research was able to train their fan fiction AI enough to halve its failure rate. Open AI will get much more than 6,000 example and they're much more motivated. They're going to do an overwhelming amount of RLHF on chatgpt3. I talk more about when AI's might lie in
Transcript
Play full episode
