> To illustrate, in the special case that GPT had a bunch of possible tokens that it judged equally probable, you could simply choose whichever token maximized g [a cryptographic function]. The choice would look uniformly random to someone who didn’t know the key, but someone who did know the key could later sum g over all n-grams and see that it was anomalously large. The general case, where the token probabilities can all be different, is a little more technical, but the basic idea is similar.

Mind blown.

A much bigger pet peeve of mine is making a SPA when a bunch of HTML pages would do, and would give you sensible URLs and the ability to open more than one tab in the first place.

When you actually need a SPA, I'd only go websockets if I really need that low latency and your clients are close enough in the first place, if they're half the way around the world on a slow connection you have different design constraints. A chat application works just fine over SSE or similar. Client-initiated requests even work with plain old fetch().

Not perfect, but it works.

I have some sympathy for this as shoplifting is getting worse. Also, an acquaintance of mine works as a security guard at a Tesco and was attacked and injured with a syringe earlier this year, luckily he's ok but they haven't caught the attacker. The cameras at the time did not have facial recognition, though I can't judge if that would have made a difference.

Thieves might ignore checkout completely, but there are also cameras capturing everyone entering and leaving the shop.