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.
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.
this isn’t me fanboying LLM corporations. pop pop pop. this article is fucking stupid though.
On Tuesday, tech stocks suffered a shock sell-off after a report from Massachusetts Institute of Technology (MIT) researchers warned that the vast majority of AI investments were yielding “zero return” for businesses.
no they didn’t. :// there was a small 1.5% “shock sell-off” (fucking lol) before rebounding. they’re only down 0.5% over the past 5-days.
even softbank, who the article focuses on, is up 36.5% (god damn) over the past month. that’s huge.
this week’s sell-off has yet to shift from a market correction to a market rout
omg stfuuuuuuuuuu. it’s -10% for a correction we aren’t even 0.5% of that.
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.
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.