Also, the engagement with "unsafe platforms" implies that schools aren't promoting the safe platform for AI or aren't effectively blocking AI on the campus. It eerily sounds like the same issues when they discovered kids could search for porn on campus.

"Misuse of AI, accidental sharing of sensitive information, or engagement with unsafe platforms can expose schools and trusts to cyber threats." is the next sentence

I'm curious to what that exact process looks like to the author. Are students sharing passwords with AI platforms???

Alternatively just in an analysis of GPT-5's performance

Keeping a record of some of the examples from the demo. The website is updated now buuut

If the real ingenuity of the 3o 4o 4.5 workflow was to be believed, it was that there was a context users could get adjusted to in order to make major productivity gains: as if different models were more adjusted or at least use LORAs. It's almost like Anthropic was right about workflows all along.

I can't help but think this was self published research at 1st glance and now am very curious about how Deakin U organizes "REDI." Blackmore is titled by the institution as a distinguished professor and the mistakes pointed out feel like they could have been spotted in the first/second draft.

If you're talking about emails and tips, that's one thing, but the implication sounds like making students "listen" to a computer which happens 0% of the time

This isn't the first slip up in standards from NYT. What happened?

If that screenshot is a manus malfunction, I'm gonna scream. 'The compounded problem of rebranding prompt engineered "GPTs" as agents is that real agents dont work' would be one hell of a subheading

is there data on actual police chases built in? I'm curious

this is a recipe for misinformation and any expert can verify by monitoring the AI while it "roleplays" because that's what it is: role-playing