Aren't you skipping a few steps between a crazy person with no background in biological lab technique simulating viruses to actually releasing something?

Can AI even simulate a virus / protein folding / etc. with any degree of accuracy? or does it just orchestrate open source tools that were already available?

In case anyone isn't aware there is also a regular non-youth version of this book that is excellent.

I was going to recommend this one! I read it at about that age and work in a cryptography-adjacent role now. Great book.

> I don't really care if someone can chain together zero-day exploits to "pwn" my lightbulbs.

The point of pwning your lightbulb is not to decide whether it's bright or dark in your house. It's valuable as a residential proxy with a good IP reputation, to be sold to the highest bidder who wants to do something they don't want traced to their own IP.

Ironically very similar to LLM output.

hmm... maybe that's where they got it from in the first place!

Sorry, which backdoor(s) exactly are you referring to?

The theory is that open models cannot follow regulations and will therefore be banned or not eligible for many of the large contracts that the frontier labs will win.

I like the part where Frog and Toad are held accountable instead of blaming the machines they built.. oh wait

Any time someone shows me something an agent did, the first question I ask is whether there's plausibly something comparable already in its training data.

OpenAI has 5 hour limits on the $20 plan. I agree about cybersecurity refusals though.

Or the industry standard will change