#AlphaFold 3

1 messages · Page 1 of 1 (latest)

vapid girder
tough bone
#

Executive summary: powerful protein/DNA structure prediction, interesting they can do two molecules interacting now. Not close to making grey goo. Maybe helpful for bioterrorists/chemical weapons manufacturing if unmonitored, but I don't think bio-weapons manufacturers usually do much structural biology anyway.

Does PauseAI have an official position on stuff like AlphaFold? I've occasionally used it as an example of relatively high benefit:risk ratio AI that wouldn't be affected by our proposed compute cap. Of course it's still an example of the need for governance and security RE aiding bad actors.

gusty mango
#

But its generally not too bad

sweet osprey
#

There are quite a few AI types that our current proposal isn't targeting at all. Pretty much every AI model actually, except for the very largest of LLMs.

I personally have some opinions on smaller, narrow AI models, but the PauseAI proposal is not focused on them

vapid girder
#

This picture perfectly sums up the general public reaction to humanity getting close to AGI

#

But back to the topic, there may be a serious conversation to be had about this kind of narrow model. I am afraid that enabling this much understanding about biology could quickly turn out to be the main risk factor