Huh, you'd think all you need are some books on the subject and some fairly generic lab equipment. Not sure what a neural net trained on Internet dumps can add to that? The information has to be in the training data for the AI to be aware of it, correct?

GPT-4 is likely trained on some data not publicly available as well.

There's also a distinction between trying to follow some broad textbook information and getting detailed feedback from an advanced conversational AI with vision and more knowledge than in a few textbooks/articles in real time.

The Tokyo Subway attack you referenced above happened in 1995 and didn't require AI. The information required can be found on the internet or in college textbooks. I suppose an "AI" in the sense of a chatbot can make it easier by summarizing these sources, but no one sufficiently motivated (and evil) would need that technology to do it.