LA Times- Say farewell to the AI bubble, and get ready for the crash
https://archive.ph/g5m2Q
As it happened, GPT-5 was a bust. It turned out to be less user-friendly and in many ways less capable than its predecessors in OpenAI’s arsenal. It made the same sort of risible errors in answering users’ prompts, was no better in math (or even worse), and not at all the advance that OpenAI and its chief executive, Sam Altman, had been talking up.
This is a bit of an oversimplification, though in fairness, I don't expect someone who isn't deeply obsessed with the AI race to understand this. Sure, GPT-5 the Product was a bust; the model router was a dud, and anyone wishing to get any good use out of GPT-5 needs to have Thinking on by default. This brings us to GPT-5 Thinking, which wasn't a total failure, being a decent improvement from 4o and a slight improvement from Grok 4. However, it's still worth noting that I still expected a little better from GPT-5, even if I wasn't expecting the same "GPT-3.5 to GPT-4" jump.
Well, not so much. When one user asked it to produce a map of the U.S. with all the states labeled, GPT-5 extruded a fantasyland, including states such as Tonnessee, Mississipo and West Wigina. Another prompted the model for a list of the first 12 presidents, with names and pictures. It only came up with nine, including presidents Gearge Washington, John Quincy Adama and Thomason Jefferson.
That is an issue with the image generation model, not the text generation model that is GPT-5.
The rest of the document, mapping a course to late 2027 when an AI agent “finally understands its own cognition,” is so loopily over the top that I wondered whether it wasn’t meant as a parody of excessive AI hype. I asked its creators if that was so, but haven’t received a reply.
I won't even dignify this with a serious response. Look, can you critics stop trying to score points and actually sit down and discuss things in a mature fashion? It's clear that the person writing this article has no understanding of how the different OAI models work.
Predictions that AI would yield a burst of increased worker productivity haven’t been fulfilled; in many fields, productivity declines, in part because workers have to be deployed to double-check AI outputs, lest their mistakes or fabrications find their way into mission-critical applications — legal briefs incorporating nonexistent precedents, medical prescriptions with life-threatening ramifications and so on.
And there are also reports of productivity improvements in other places. It's not "one or the other."
