GPT5 just proved what many of us in the software industry on the technical side have been saying since the beginning.
LLMs are not AI. And they are only as useful as the information they are trained on, and with the industry using all of the internet to train the majority of them, they have tons of false information. And everything an LLM says is based on s confidence level that it’s correct, but those confidence levels are configured so low that it’s often way off. Plus people are used to computers giving specific, correct answers, but LLMs are all about small talk and making things up to fill time because that’s what they’re trained on. People need to learn these aren’t AI and everything they say needs to be taken with that in mind. Double and triple check it before believing it. But since we often don’t even do that with humans, thus the whole anti-vax thing for example, it’s even harder for people to want to do that for something that was explicitly marketed as preventing them from needing to do the research.
So now that those not profiting directly from AI are seeing that it’s probably never going to get better as promised, they’re losing confidence. But it will stick around for s while. The energy industry is powerful and is lobbying hard for it since it’s the first time in a while our energy demands have spiked so high. And with the mechanisms to monitor climate change being shut down or explicitly destroyed in the US, and conservatives convinced that the natural disasters are just short term, dirty fuel demand is back, more than ever. So they have a ton of incentive to keep it alive ss well as the companies making it.
Thorry84@feddit.nl 2 days ago
It’s because GPT5 was such a disappointment.
Partly because they hyped it up beyond belief and partly because the results are actually bad. GPT3 to 4 was a big step up and with all the marketing people expected 4 to 5 to be just as big of a step. It turned out to be mostly a step sideways and in many aspects even a step back. Energy consumption is rumoured to be up, hallucinations are up and user experience is way down. At the same time all of the very public and high profile promises made are being broken and for a change people actually noticed.
The sentiment in the market was down for a while, but GPT5 release really kicked everything into high gear. Earlier people were disapointed “AI” didn’t live up to the hype. Regular folks use it all the time as a replacement search engine, because Google had gotten so bad. But in businesses adoption was slow and where it was used the gains promised weren’t seen. If the product was given away for free, people would use it, but even modestly paid subscriptions weren’t taking off. But people thought: Hey, this is just the current level of tech, the next level is going to be so much better and improve a lot right? GPT5 proved that wasn’t really true, so people lost faith fast.
LLM system are rapidly running into diminishing returns with larger models not yielding better results and sometimes even worse results. They also run into the issue they’ve poisoned the well. They used to train on data from the internet, but with the internet being flooded by output from older AI models, it’s eating it’s own shit. That’s really bad for the performance of the newer models. Especially on things like coding and such, where it relies on code examples to produce new code. With Stackoverflow dying due to a lot of things, but AI being a big factor, there isn’t as much of that as there used to be. Same for other stuff on the internet, once you kill the internet that great source of data is lost.
Now I don’t think it’s as worse (or good depending on your point of view) as some articles make it out to be. Many companies still see AI as the infinite money well the marketing claims it to be. A lot of people use it all the time, even though they’ve had negative experiences with it. But it’s somewhat good to see some reality seeping in here and there.
I fully expect the shit to hit the fan and the bubble to burst in a catastrophic way. This will be bad, way worse than when the dotcom bubble burst. It’s not going to be good for anyone. But better it burst soon than keep pumping it up further.
MCasq_qsaCJ_234@lemmy.zip 2 days ago
Well, GPT-5 was a disappointment because it didn’t live up to expectations, but will it really cause the bubble to burst?
Because user frogbellyratbone_ gave a good explanation.