> We've been hearing about the risks of engineered viruses and homemade superweapons since GPT 3.5, so where are they? We've had abliterated open weights models much stronger than GPT 4 for over a year now.
replyWhat I can't understand, is that they act like the _knowledge_ is dangerous.
I don't know if I'm biased from my BSci (chem/maths), but: knowledge isn't dangerous, the reagents needed are incredibly easy to control. Thats what we already do!
They have existed far longer than LLms, and kill tens of thousands of Americans every year.
Fentanyl is the best example. Any chemistry graduate can figure out how to manufacture it. Used by China as a Chemical weapon against the American public.
replyWhat is this comment? If they occurred we would face a huge disaster; isn't it better to err on the side of caution to make that risks as low as possible???
replyWhat are you willing to give up to be less afraid of all these extremely hypothetical risks? (which in the case of bio/nuclear/chemical is mostly access to controlled materials and capital anyway)
replyAccess to information and knowledge is probably the very last thing I'd be willing to give up, personally.
I think most people would be a lot happier if they were less fearful in general.
No no we should push the limits until a bioattack happens, then when those people are all dead we comment angrily on the hackernews thread and say that someone should have seen this coming
replyMy smoke detector has gone off three times now, where is the fire?
reply>We've been hearing about the risks of engineered viruses and homemade superweapons since GPT 3.5, so where are they? We've had abliterated open weights models much stronger than GPT 4 for over a year now.
replyTry...since GPT 2.
Come on, no one was worried that GPT-2 would help people engineer viruses. The concern was generating misinformation and spam.
reply