The essential read on why the flop of GPT5 means so much more than AI companies admit and could be a wake-up-call for all who bought into the hype. Yes, Gary Marcus was right.

Aug 10, 2025 · 8:43 AM UTC

17
152
20
1,043
Gary Marcus - bookmark his overview of the failure of Chat GPT 5 as one of historic significance. garymarcus.substack.com/p/gp…
This simple essay has potentially worldwide economic implications. It means the AI market bubble will pop soon because $500bill was borrowed and squandered on a South Sea-like venture.
2
4
27
Yes, this essay needs to be shared.
10
Replying to @MrEwanMorrison
I love these articles. They give me hope. 🤏
1
13
He's been telling the truth for years. Now people are really listening.
1
20
Haven’t heard the ‘these models will just continue to get better’ line in a while. They must be on to a new marketing strategy.
2
8
The usual AI pushers are just being silent for a little while and hoping no-one will remember they ever said "these AI models will just continue to get better" "AGI is solved".
1
5
19
Replying to @MrEwanMorrison
The problem with some startups and hype cycles is that there tends to be some that over hype and create a huge bubble. Like a pump and dump scheme. Someone is going to get rich and then many will be caught in the wake of disaster.
1
1
Will Open AI survive the crash in confidence - as you say there has to be a sacrificial pump and dump
1
Replying to @MrEwanMorrison
Makes a lot of sense. I didn't know about some of those things in this article, but figured there'd be limits. That's totally fine too. It just means more algorithms and tools may be needed for the task at hand (depending on the task). That keeps people busy and employed. Good.
You're missing the bigger point and saying "all this mess can be fixed by the same tools that created this mess"
1
1
Replying to @MrEwanMorrison
The only predictable thing about AI is Gary Marcus being wrong, and generally missing the point.
2
2
You haven't been keeping up to date on X over the last few days.
1
3
Replying to @MrEwanMorrison
the hype train was running off the rails. people need to step back and see the bigger picture instead of just chasing the latest shiny toy.
Replying to @MrEwanMorrison
lol no
No courage, low product design skill Deprecate them all, leave gpt5 only. Think on by default Don’t even have 2 model options when it’s really just one as per api. User can prompt for it explicitly or model is smart enough to know 0 to N thinking tokens needed for the prompt
Replying to @MrEwanMorrison
He's retarded
1
Replying to @MrEwanMorrison
FlopGPT — I like that 😆👏
1
Replying to @MrEwanMorrison
I don’t think GPT-5 can be called a flop. OpenAI did a bang-up job. The 192K context, multi-agent workflows, and better reasoning make it a clear leap forward for a lot of real-world use cases, even if the new ChatGPT didn’t match everyone’s hype cycle fantasies.
Replying to @MrEwanMorrison
I used the information in this thread to join the class action, and listed all 7 of my books that were taken illegally 🌈the more you know🌈