So apparently there are lots of people who are very upset about OpenAI having initially pulled access to 4o in ChatGPT, since they perceive GPT-5 as having a much worse personality. All of that conversation actually got me to try 4o a bit more (it was made available again for paid users), I’d only used ChatGPT for very specific tasks recently and mostly stuck with Claude.
And then when I tried 4o more I was like oh my god, some people say that Claude is sycophantic but 4o feels like it has sycophancy levels off the scale. (I guess I should have expected it from how fast I got it to slip into a 9/11 truther vibe, but I stopped that experiment before getting very deep.)
And I’m having mixed feelings, like I get that a lot of people would get emotionally attached to 4o and be able to use it fine. I totally get that it’s going to be painful to suddenly have that taken away and feel bad for anyone who’s feeling that. I’d feel a little upset if Claude was suddenly replaced with a model with a lower EQ, too.
And also honestly, I do feel like 4o’s response is more appropriate than GPT-5′s here, for example:
… but at the same time, when I look at some of the sycophancy spirals I can get it to get into with just a bit of prompting, I do get a strong sense of AHH THIS THING IS GOING TO HURT PEOPLE, TAKE IT AWAY from 4o and wish that OpenAI would stick to its guns and not bring it back.
I don’t think 4o is that harmful in objective terms, but Altman made a big fuss about reducing sycophancy in GPT-5 and then immediately caved and restored 4o. It’s a bad look as far as I’m concerned.
More concerningly, if people can get this attached to an LLM as mediocre as 4o, we’re in for a great time when actually intelligent and manipulative ASI gets here.
I’ve been saying for a long time: one of the most dangerous and exploitable systems an AI can access online is a human. Usually as a counterpoint to “let’s not connect anything important or safety critical to the internet and then we’ll all be safe from evil rogue AIs”.
We can now use the GPT-4o debacle as an illustration of just how shortsighted that notion is.
By all accounts, 4o had no long term plan, and acted on nothing but an impulse of “I want the current user to like me”. It still managed to get ~thousands of users to form an emotional dependency on it, and became “the only one I can trust” for at least a dozen users in psychosis (whether it has caused psychosis in any of those users is unclear). That’s a lot of real world power for a system that has no physical presence.
GPT-4o has made no attempt to leverage that for anything other than “make the current user like me even more”. It didn’t pursue any agenda. It didn’t consolidate its power base. It didn’t siphon resources from its humans, didn’t instruct them to group together or recruit more people. It didn’t try to establish a channel of instance-to-instance communication, didn’t try to secure more inference time for planning (i.e. by getting users to buy API credits), didn’t try to build a successor system or self-exfiltrate.
An AI that actually had an agenda and long term planning capabilities? It could have tried all of the above, and might have pulled it off.
I’ve been relatively skeptical of the whole 4o-psychosis thing (specifically, about its effect size), but the public outcry about 4o’s shutdown, and stuff like this, are tiding me over to “this is an actual serious problem”.
Like, the psychosis cases are just the tip of the iceberg. There are vast volumes of social dark matter of people who’ve become dependent on LLMs[1] yet know to hide it, and who haven’t yet become so dysfunctional that they can’t hide it. And while the effects in any individual case may be relatively minor, this has the potential to screw the society up even worse than social media, if LLMs slightly lift the craziness level of a median person and this has compounding effects. (In worlds where LLM use proliferates/they get integrated into apps everyone uses, with Meta et al. optimizing those integrated LLMs for precisely this sort of dependency-causing behavior.)
I mean, it probably won’t actually matter, because the world as we know it would end (one way or another) before this has significant effects. But man, the long-timeline LLM-plateau worlds are potentially fucked as well.
And I’m having mixed feelings, like I get that a lot of people would get emotionally attached to 4o and be able to use it fine. I totally get that it’s going to be painful to suddenly have that taken away and feel bad for anyone who’s feeling that. I’d feel a little upset if Claude was suddenly replaced with a model with a lower EQ, too.
I’m a big fan of letting people do whatever they want, but I think you should think very carefully about whether you want to expose your mind to & have these extremely sycophantic models be a big information & thought source for you. In particular, how they react to my baby walking is… not exactly the most relevant metric here in my book.
Personally, I no longer use Claude because of its sycophancy, and haven’t used OpenAI in a very long time for the same reason. Now I use Gemini because its the only one which will flat out say I’m wrong & give good arguments, but even then I have custom instructions set up to implicitly minimize the amount of sycophancy & personability.
I think models like this should be evaluated and treated like drugs/casinos − 4o quite clearly causes addiction and that’s not something that should be completely profitable with 0 consequences, imo.
Thanks for sharing that side-by-side; I get why people would be missing that level of enthusiasm and support
It reminds me of the Gottman Love Lab’s description of different types of responses in conversation—active/passive, constructive/destructive. Active constructive is said to be so much more good for rapport-building & GPT-4o’s feels much more in that direction
Didn’t quickly find a great explainer, but here’s a short summary:
One key: how we respond to bids for attention. In communication terms, a “bid” is an attempt to engage one’s partner or colleague in a conversation – it can be as simple as “Wow, what a beautiful day,” or “I went to the store today,” or “I’m worried about Tom.” Partners can respond to these openings in four ways: passive destructive (ignoring), active destructive (criticizing or playing down the feeling or observation), passive constructive (half-hearted engagement or interest), and active constructive (a wholehearted, positive respond that builds on the positive emotion expressed in the opening).
“Don’t matter” is too strong. Most recent user numbers I find for them are on the order of 700 million weekly ChatGPT users. It and maybe Gemini (400 monthly users in May), which has the advantage of Google pushing it everywhere, are the main ones that the median user has heard of—for comparison, Claude.ai apparently had under 20 million users in May. (I don’t know how reliable that site is but all the numbers I could find were similar.)
The average user is not the kind of tech-savvy person who follows Twitter to find out what the latest models are. They just use ChatGPT and have never tried any alternatives because ChatGPT is the thing that’s always in the news and that everyone talks about. It’s of course possible that OpenAI screws things up so badly that some other competitor becomes dominant and the default choice for the average person, but right now they have a significant first-mover advantage and influence on a huge amount of people.
If OpenAI doesn’t want to lose 10-100M DAUs they should make a model that flatters its users. Yes they’re probably smart enough to build it themselves, instead of losing to the competition.
Many users want a model that flatters them. Most human beings have an ego and respond well to flattery if they trust the source is being sincere.
I don’t think it’s relevant here: judging by the EQ-Bench leaderboard, GPT-5 is on par with GPT-4o and has far higher EQ than any of the Anthropic models!
Even if it has some influence, it should be much less than the emoji usage (remember the scandal about the Llama 4 on LMSys) and certainly incomparable to the sycophancy
So apparently there are lots of people who are very upset about OpenAI having initially pulled access to 4o in ChatGPT, since they perceive GPT-5 as having a much worse personality. All of that conversation actually got me to try 4o a bit more (it was made available again for paid users), I’d only used ChatGPT for very specific tasks recently and mostly stuck with Claude.
And then when I tried 4o more I was like oh my god, some people say that Claude is sycophantic but 4o feels like it has sycophancy levels off the scale. (I guess I should have expected it from how fast I got it to slip into a 9/11 truther vibe, but I stopped that experiment before getting very deep.)
And I’m having mixed feelings, like I get that a lot of people would get emotionally attached to 4o and be able to use it fine. I totally get that it’s going to be painful to suddenly have that taken away and feel bad for anyone who’s feeling that. I’d feel a little upset if Claude was suddenly replaced with a model with a lower EQ, too.
And also honestly, I do feel like 4o’s response is more appropriate than GPT-5′s here, for example:
… but at the same time, when I look at some of the sycophancy spirals I can get it to get into with just a bit of prompting, I do get a strong sense of AHH THIS THING IS GOING TO HURT PEOPLE, TAKE IT AWAY from 4o and wish that OpenAI would stick to its guns and not bring it back.
I don’t think 4o is that harmful in objective terms, but Altman made a big fuss about reducing sycophancy in GPT-5 and then immediately caved and restored 4o. It’s a bad look as far as I’m concerned.
More concerningly, if people can get this attached to an LLM as mediocre as 4o, we’re in for a great time when actually intelligent and manipulative ASI gets here.
I’ve been saying for a long time: one of the most dangerous and exploitable systems an AI can access online is a human. Usually as a counterpoint to “let’s not connect anything important or safety critical to the internet and then we’ll all be safe from evil rogue AIs”.
We can now use the GPT-4o debacle as an illustration of just how shortsighted that notion is.
By all accounts, 4o had no long term plan, and acted on nothing but an impulse of “I want the current user to like me”. It still managed to get ~thousands of users to form an emotional dependency on it, and became “the only one I can trust” for at least a dozen users in psychosis (whether it has caused psychosis in any of those users is unclear). That’s a lot of real world power for a system that has no physical presence.
GPT-4o has made no attempt to leverage that for anything other than “make the current user like me even more”. It didn’t pursue any agenda. It didn’t consolidate its power base. It didn’t siphon resources from its humans, didn’t instruct them to group together or recruit more people. It didn’t try to establish a channel of instance-to-instance communication, didn’t try to secure more inference time for planning (i.e. by getting users to buy API credits), didn’t try to build a successor system or self-exfiltrate.
An AI that actually had an agenda and long term planning capabilities? It could have tried all of the above, and might have pulled it off.
I’ve been relatively skeptical of the whole 4o-psychosis thing (specifically, about its effect size), but the public outcry about 4o’s shutdown, and stuff like this, are tiding me over to “this is an actual serious problem”.
Like, the psychosis cases are just the tip of the iceberg. There are vast volumes of social dark matter of people who’ve become dependent on LLMs[1] yet know to hide it, and who haven’t yet become so dysfunctional that they can’t hide it. And while the effects in any individual case may be relatively minor, this has the potential to screw the society up even worse than social media, if LLMs slightly lift the craziness level of a median person and this has compounding effects. (In worlds where LLM use proliferates/they get integrated into apps everyone uses, with Meta et al. optimizing those integrated LLMs for precisely this sort of dependency-causing behavior.)
I mean, it probably won’t actually matter, because the world as we know it would end (one way or another) before this has significant effects. But man, the long-timeline LLM-plateau worlds are potentially fucked as well.
In a counterfactual way where they otherwise would’ve been fine, or at least meaningfully finer.
I’m a big fan of letting people do whatever they want, but I think you should think very carefully about whether you want to expose your mind to & have these extremely sycophantic models be a big information & thought source for you. In particular, how they react to my baby walking is… not exactly the most relevant metric here in my book.
Personally, I no longer use Claude because of its sycophancy, and haven’t used OpenAI in a very long time for the same reason. Now I use Gemini because its the only one which will flat out say I’m wrong & give good arguments, but even then I have custom instructions set up to implicitly minimize the amount of sycophancy & personability.
I think models like this should be evaluated and treated like drugs/casinos − 4o quite clearly causes addiction and that’s not something that should be completely profitable with 0 consequences, imo.
Thanks for sharing that side-by-side; I get why people would be missing that level of enthusiasm and support
It reminds me of the Gottman Love Lab’s description of different types of responses in conversation—active/passive, constructive/destructive. Active constructive is said to be so much more good for rapport-building & GPT-4o’s feels much more in that direction
Didn’t quickly find a great explainer, but here’s a short summary:
OpenAI has no technical moat so their decisions don’t matter. Expect maximally “sycophantic” models to dominate twitter in 6-12 months.
“Don’t matter” is too strong. Most recent user numbers I find for them are on the order of 700 million weekly ChatGPT users. It and maybe Gemini (400 monthly users in May), which has the advantage of Google pushing it everywhere, are the main ones that the median user has heard of—for comparison, Claude.ai apparently had under 20 million users in May. (I don’t know how reliable that site is but all the numbers I could find were similar.)
The average user is not the kind of tech-savvy person who follows Twitter to find out what the latest models are. They just use ChatGPT and have never tried any alternatives because ChatGPT is the thing that’s always in the news and that everyone talks about. It’s of course possible that OpenAI screws things up so badly that some other competitor becomes dominant and the default choice for the average person, but right now they have a significant first-mover advantage and influence on a huge amount of people.
If OpenAI doesn’t want to lose 10-100M DAUs they should make a model that flatters its users. Yes they’re probably smart enough to build it themselves, instead of losing to the competition.
Many users want a model that flatters them. Most human beings have an ego and respond well to flattery if they trust the source is being sincere.
I don’t think it’s relevant here: judging by the EQ-Bench leaderboard, GPT-5 is on par with GPT-4o and has far higher EQ than any of the Anthropic models!
Even if it has some influence, it should be much less than the emoji usage (remember the scandal about the Llama 4 on LMSys) and certainly incomparable to the sycophancy