Connor Leahy on AI alignment

I have a hard time keeping up with AI developments, It is the beginning of March and it feels like there have already been more breakthroughs than what we have seen in all to 2023. There are billions and billions flowing into AI right now.

Google Gemini suffered a quite humiliating over-guiderail issue where it refused to generate white people. Also on the text generation side there is proof of some over censoring trying to guide the models to adhere to some kind of moral. ethical target.

If this is the state of alignment, we may indeed have a problem. Connors worries and views should be listened to. More dialog is needed.