I still think GPT5 is really a cost cutting measure, with a company that is trying to grow to 1 billion users on a product that needs GPUs.
I dont see anyone talking about GPT 5 Pro, which I personally tested against:
- Grok 4 Heavy
- Opus 4.1
It was far better than both of those, and is completely state of the art.
The real story is running these models at true performance max likely could go into the thousands per month per user. And so we are being constrained. OpenAI isnt going for that market segment, they are going for growth to take on Google.
This article doesnt have one reference to the Pro model. Completely invalidates this guys opinion
Can someone remind me of anything Gary has contributed to AI?
Last I saw he hasn’t produced anything but general “pop” books on AI and being associated with MIT, which IMO has zero weight on applied or even at this point theoretical AI, as that is primarily coming out of corporate labs.
No new algorithms, frameworks, datasets, products, insights.
Why is this guy relevant enough to keep giving him attention, his entire ouvre is just anti-whatever is getting attention in the “AI” landscape
I don’t see him commenting on any other papers and he has no lab or anything
Someone make it make sense, or is it as simple as “yeah thing makes me feel bad, me repost, me repeat!”
Yeah, the sycophancy withdrawal is real. I almost considered telling GPT-5 to act ten years younger, use emoji everywhere, and compliment me at the beginning of every response... but I snapped out of it.
The AI community requires more independent experts like Marcus to maintain integrity and transparency, ensuring that the field does not succumb to hyperbole as well as shifting standards such as "internally achieved AGI", etc.
Regardless of personal opinions about his style, Marcus has been proven correct on several fronts, including the diminishing returns of scaling laws and the lack of true reasoning (out of distribution generalizability) in LLM-type AI.
These are issues that the industry initially denied, only to (years) later acknowledge them as their "own recent discoveries" as soon as they had something new to sell (chain-of-thought approach, RL-based LLM, tbc.).
GPT-5 is just OpenAI getting started. Just wait and see what GPT-6 is capable of and imagine that GTP-6 is just OpenAI getting started: if GPT-6 was a high school student, GPT-7 is an expert with masters degree; but GPT-7 is OpenAI getting started
This is a genre of article I find particularly annoying. Instead of writing an essay on why he personally thinks GPT-5 is bad based on his own analysis, the author just gathers up a bunch of social media reactions and tells us about them, characterizing every criticism as “devastating” or a “slam”, and then hopes that the combined weight of these overtorqued summaries will convince us to see things his way.
It’s both too slanted to be journalism, but not original enough to be analysis.
People on our circles are obsessed with model performance. OpenAI's lead is not there and hasn't been there for some time.
They do, however, have a major lead in terms of consumer adoption. To normal people who use llm's, ChatGPT is _the_ model.
This gives them a lot of opportunities. I don't know what's taking them so long to launch their own _real_ app store, but that's the game they are ahead of everyone else because of the consumer adoption.
I swear I had an understanding of how to get deep analytical thinking out of o3. I am absolutely struggling to get the same results with GPT-5. The new model feels different and frustrating to use.
A friend of mine works in AI professionally. He told me months ago that it is basically just all a scam and hype to garner investment money. He said the technology and paradigm itself will never lend toward AGI or anything like that.
"People had grown to expect miracles, but GPT-5 is just the latest incremental advance."
This is really the only part of the article I think was worth writing.
-People should expect an incremental advance
-Providers should not promise miracles
Managing expectations is important. The incremental advances are still advances, though, even if I don't think "AGI" is just further down on the GPT trajectory.
same sentiments with an article author - gpt5 looks like a cost-cut initiative.
my personal feeling gpt5-thinking is much faster but doesnt produce the same quality results as o3 which were capable to scan through the code base dump with file names and make correct calls
Can someone let me know if they also find the existing UI prompt unbearably slow? At first I thought it was my browsers but I am having the same experience on every machine. It's so bloody slow with loading responses, freezing and even giving me the old browser tab death warning "Cancel or Wait"
There is no training data left. Every refinement in AI from here on will come from architectural changes. All models basically have reached a local maximum on new information.
Gary Marcus would have wrote this article in all possible scenarios, unless ChatGPT 5 was literally AGI (maybe even it were, he would still have found something to attack). There is valid criticism, and there is just being a contrarian for the sake of being a contrarian.
The whole thing feels less like “Hey, this is why I think the model is bad” and more like the kind of sensationalist headline you’d read in a really trashy tabloid, something like: “ChatGPT 5 is Hot Garbage, Totally Fails, Sam Altman Crushed Beneath His Own Failure.”
Also, I have no idea why people give so much attention to what this guy has to say.
Here are my reasons why this "upgrade" is, in experience, a huge downgrade for Plus users:
* The quality of responses from GPT-5 compared to O3 is lacking. It does very few rounds of thinking and doesn't use web search as O3 used to. I've tried selecting "thinking", instructing explicitly, nothing helps. For now, I have to use Gemini to get similar quality of outputs.
* Somehow, custom GPTs [1] are now broken as well. My custom grammar-checking GPT is ignoring all instructions, regardless of the selected model.
* Deep research (I'm well within the limit still) is broken. Selecting it as an option doesn't help, the model just keeps responding as usual, even if it's explicitly instructed to use deep research.
63 comments
[ 2.4 ms ] story [ 59.9 ms ] threadI dont see anyone talking about GPT 5 Pro, which I personally tested against:
- Grok 4 Heavy
- Opus 4.1
It was far better than both of those, and is completely state of the art.
The real story is running these models at true performance max likely could go into the thousands per month per user. And so we are being constrained. OpenAI isnt going for that market segment, they are going for growth to take on Google.
This article doesnt have one reference to the Pro model. Completely invalidates this guys opinion
Last I saw he hasn’t produced anything but general “pop” books on AI and being associated with MIT, which IMO has zero weight on applied or even at this point theoretical AI, as that is primarily coming out of corporate labs.
No new algorithms, frameworks, datasets, products, insights.
Why is this guy relevant enough to keep giving him attention, his entire ouvre is just anti-whatever is getting attention in the “AI” landscape
I don’t see him commenting on any other papers and he has no lab or anything
Someone make it make sense, or is it as simple as “yeah thing makes me feel bad, me repost, me repeat!”
Regardless of personal opinions about his style, Marcus has been proven correct on several fronts, including the diminishing returns of scaling laws and the lack of true reasoning (out of distribution generalizability) in LLM-type AI.
These are issues that the industry initially denied, only to (years) later acknowledge them as their "own recent discoveries" as soon as they had something new to sell (chain-of-thought approach, RL-based LLM, tbc.).
It’s both too slanted to be journalism, but not original enough to be analysis.
They do, however, have a major lead in terms of consumer adoption. To normal people who use llm's, ChatGPT is _the_ model.
This gives them a lot of opportunities. I don't know what's taking them so long to launch their own _real_ app store, but that's the game they are ahead of everyone else because of the consumer adoption.
I know AI hype is truly insane, but surely nobody actually believed the singularity was upon us?
It seems to lose the thread of the conversation quite abruptly, not really knowing how to answer the next comment in a thread of comments.
It's like there is some context cleanup process going on and it's not summarizing the highlights of the conversation to that point.
If that is so, then it seems to also have a very small context, because it seems to happen regularly.
Asking it to 'Please review the recent conversation before continuing' prompt seems to help it a bit.
I swear I had an understanding of how to get deep analytical thinking out of o3. I am absolutely struggling to get the same results with GPT-5. The new model feels different and frustrating to use.
He sent me all these articles geared toward that end as well. https://garymarcus.substack.com/p/seven-replies-to-the-viral... https://substack.com/@cattelainf/note/c-135021342 https://arxiv.org/abs/2002.06177 https://garymarcus.substack.com/p/the-ai-2027-scenario-how-r... https://garymarcus.substack.com/p/25-ai-predictions-for-2025...
This is really the only part of the article I think was worth writing.
-People should expect an incremental advance
-Providers should not promise miracles
Managing expectations is important. The incremental advances are still advances, though, even if I don't think "AGI" is just further down on the GPT trajectory.
my personal feeling gpt5-thinking is much faster but doesnt produce the same quality results as o3 which were capable to scan through the code base dump with file names and make correct calls
dont feel any changes with https://chatgpt.com/codex/
my best experience was to use o3 for task analysis, copy paste the result in https://chatgpt.com/codex/, work outside and vibe code from mobile
Gpt5 was an incremental improvement. That’s fine. Was hyped hard but what did you expect? It’s part of the game
Meanwhile the fraction of text that the real world consists of is microscopic.
The whole thing feels less like “Hey, this is why I think the model is bad” and more like the kind of sensationalist headline you’d read in a really trashy tabloid, something like: “ChatGPT 5 is Hot Garbage, Totally Fails, Sam Altman Crushed Beneath His Own Failure.”
Also, I have no idea why people give so much attention to what this guy has to say.
* The quality of responses from GPT-5 compared to O3 is lacking. It does very few rounds of thinking and doesn't use web search as O3 used to. I've tried selecting "thinking", instructing explicitly, nothing helps. For now, I have to use Gemini to get similar quality of outputs.
* Somehow, custom GPTs [1] are now broken as well. My custom grammar-checking GPT is ignoring all instructions, regardless of the selected model.
* Deep research (I'm well within the limit still) is broken. Selecting it as an option doesn't help, the model just keeps responding as usual, even if it's explicitly instructed to use deep research.
[1]: https://openai.com/index/introducing-gpts/