Rob Wiblin

Among others, you're gonna suggest, for example, that we should aim to have like broad spectrum antivirals, broad spectrum vaccines that can potentially give us immunity to an entire class of viruses rather than just like an individual strain.

But as the technology advances further, you might imagine that a really hostile actor that also has access to these incredible AI biology design tools is going to be able to cherry pick the exact strain of virus, be able to design something that can evade the known vaccines that you've distributed to your population trying to defend them.

So what would you say to someone, perhaps me, or I guess another pessimistic feeling listener who might feel like, why should I stick around to hear the rest of this?

This just seems like such a difficult situation.

So hopeless that I just think that we're not going to be able to make a real dent on the size of the risk.

Medicine.

Okay, let's talk about access controls first.

What is worth doing on access controls?

And do you agree with my take that it's kind of more of a delaying tactic than anything else?

A really relevant audience question that was repeated by multiple people is, which do you think is generating more of the risk, open source models or closed source models?

So it depends on the actor.

So to sum that up, is it that for amateur actors or novice actors, the open weight models are enough to help them because they need help with more basic stuff.

They don't need to be doing cutting-edge science.

For the much more capable actors, like as bioweapons programs, you know...

academics in the field, it's the very best models that are actually going to be moving the needle for them.

I mean, I guess they could work faster with an open source model that's somewhat less impressive.

But I suppose you think in terms of the risk generated, what you'd be most worried about with those folks is that they come up with something that's just radically more lethal or radically better at spreading, something that hasn't been done before, basically.

Okay, so come back to it.

What are your priority asks on access controls?

What do you want the companies or governments to be doing?