One has to wonder if one of the main innovations driving "AI" is the complete lack of accountability and even shame.

Twenty years ago, we wouldn't have had companies framing the raw output of a text generator as some kind of complete product, especially an all-encompassing general one. How do you know that these probabilistic text generators are performing valid synthesis, as opposed to word salad? You don't. So LLM technology would have used to do things like augment search/retrieval, pointing to concrete sources and excerpts. Or to analyze a problem using math, drive formal models that might miss the mark but at least wouldn't be blatantly incorrect with a convincing narrative. Some actual vision of an opinionated product that wasn't just dumping the output and calling it a day.

Also twenty years ago we also wouldn't have had a company placing a new beta-quality product (at best) front and center as a replacement for their already wildly successful product. But it feels like the real knack of these probabilistic word generators is convincing "product people" of their supreme utility. Of course they're worried - they found something that can bullshit better than themselves.

At any rate all of those discussions about whether humans are be capable of keeping a superintelligent AI "boxed" are laughable in retrospect. We're propping open the doors and chumming other humans' lives as chunks of raw meat, trying to coax it out.

(Definitely starting to feel like an old man here. But I've been yelling at Cloud for years so I guess that tracks)