Rob Wiblin

They then actually made a bunch of those viruses and found that quite a lot of them were viable.

Tell us more about that experiment.

What are some other empirical results we've had in recent years that convince you that there is a real AI biosecurity problem to be solved?

Because it's created by a bacteria?

So the ultimate bad scenario here would be if terrorists could basically go to an AI model and say, here's this known terrible pathogen for humans, but I can't get this synthesized because it would be screened out and I would be flagged.

So can you please change as many amino acids here?

Please change as much of the DNA as you can without basically changing the shape of the thing at all.

So it basically functions identically, but it's not going to be picked up at any point as something that is dangerous.

Exactly.

And I guess we're not there yet because I suppose they were only changing a single protein to change lots of proteins through an entire genome and not mess one of them up.

That's a higher bar.

But these things are always improving.

Yeah, you said there was a different empirical result that troubled you, which is focused on human language models in particular.

What's that?

Yeah, it might be worth giving people some context that I guess for a long time people have been worried that terrorists or bad actors or rogue countries might be able to develop new biological weapons or new pandemics that we really wouldn't like.

And isn't science advancing?

All of these tools are increasing.

Knowledge is disseminating.

That should make us very worried.

And probably the most intelligent, the most reasonable response has been it's not enough to have a bunch of textbook knowledge, like explicit knowledge that you could Google or look up in a virology textbook.