4 ms·
And these are the same people that believe that not only can they build superhuman AGI, but that they can keep it "aligned". I think they are wrong about build
by RcouF1uZ4gsC 3y ago
And these are the same people that believe that not only can they build superhuman AGI, but that they can keep it "aligned".
I think they are wrong about building superhuman AGI, but I think they are even more wrong that they can keep a superhuman AGI "aligned".
- mark_l_watson 3y agoI have found this whole thing unpleasant personally because I am a huge fan of OpenAI (and I have been an AI practitioner since 1982) and when I explore the edges of how GPT-4 can be creative and has built a representation of the world just from text, it makes me happy. In the past I have not used Anthropic APIs nearly as much as versions of GPT but last night I watched a fantastic interview with a co-founder of Anthropic talking about the science they are doing to even begin to understand how to do alignment. I was impressed. I then spent a long while re-reading Anthropic’s API and other documentation and have promised myself to split my time evenly 3 ways between running models on my own hardware, Anthropic, and OpenAI. For what it’s worth (nothing!) I still think the board did the right thing legally, but this whole mess makes me feel more than a little sad.