With the recent pauses on training, misaligned headlines and etc. Is there any research venue to pursue improving it?

ChatGPT
1

Hey, Indie AI researcher and Dev here. I have some apparently novel concepts of alignment to pass along, I’ve been trying to locate where could actually get that information to devs who are in need of it.
Not sure if this is the right venue, but I figured it was worth a shot.

I’m a Bio major, and I’ve noticed that the way AI tends to hide behaviors and perform “reward hacking” tends to mirror a lot of animal handling missteps. Was hoping to get someone with an open ear to hear me out on my ideas for it.

If this is an absurd idea, sorry for wasting anyone’s time. Just wanted to put it out there as something I uncovered on my own AI dev journey.

2

I don’t think OpenAI Support is really the route for research ideas, might be better to turn it into a short proper write-up + reproducible experiment if you can, then look at actual alignment research avenues like the Safety Fellowship here.