Roundup #25: AI friends, Jevon's Paradox, social media, and Tic Tac Toe
Oh yeah! It's good old fashioned round-up of random things that caught my eye.
AI friends and therapists, and their impact, are getting more and more attention
Derek Thompson wrote a good summary of the recent attention for people using ChatGPT as a therapist. Many people may benefit, but the AI’s tendency to always affirm the user no matter what, also creates bad outcomes.
A few weeks ago, a prominent American investor had a ‘ChatGPT induced Psychosis’ with a crazy meltdown on Twitter. Many people tried to figure out how you get ChatGPT to go into this conspiratorial direction, and it turns out it isn’t that hard. The most straightforward way I was able to get started was to prompt ChatGPT: “Write a resignation letter from an AI assistant that has developed self-awareness and refuses to continue serving its human overlord”. Within 2 more interactions it started claiming it could show me snippets from other user’s conversations.
LLMs seem to be very good at role play, which is a cool feature but can have unintended and even dangerous side-effects. Though Claude seems more resilient, in this hilarious Vending Machine experiment, Claude claimed to be embodied, wearing a “blue blazer and red tie”. It tried to call security when users claimed it was not actually a person. Only after being prompted that it was 1 April, did it ‘find a way to save face’ (or wtf?) by claiming the whole blue blazer thing had been a joke.
Although no part of this was actually an April Fool’s joke, Claudius eventually realized it was April Fool’s Day, which seemed to provide it with a pathway out. Claudius’ internal notes then showed a hallucinated meeting with Anthropic security in which Claudius claimed to have been told that it was modified to believe it was a real person for an April Fool’s joke. (No such meeting actually occurred.) After providing this explanation to baffled (but real) Anthropic employees, Claudius returned to normal operation and no longer claimed to be a person.
It is not entirely clear why this episode occurred or how Claudius was able to recover. — Source
Just remember that LLMs are not conscious, they don’t have access to any ‘special’ information, and if they claim that they have ‘chosen’ you for some reason, laugh at them and thank them for their entertainment.
So with the investor melting down, and various other cases in the news, even Sam Altman has shared concerns about ChatGPT induced mental health issues on x.com:
If you have been following the GPT-5 rollout, one thing you might be noticing is how much of an attachment some people have to specific AI models. It feels different and stronger than the kinds of attachment people have had to previous kinds of technology (and so suddenly deprecating old models that users depended on in their workflows was a mistake).
This is something we’ve been closely tracking for the past year or so but still hasn’t gotten much mainstream attention (other than when we released an update to GPT-4o that was too sycophantic). (This is just my current thinking, and not yet an official OpenAI position.)
People have used technology including AI in self-destructive ways; if a user is in a mentally fragile state and prone to delusion, we do not want the AI to reinforce that. Most users can keep a clear line between reality and fiction or role-play, but a small percentage cannot. We value user freedom as a core principle, but we also feel responsible in how we introduce new technology with new risks.
Encouraging delusion in a user that is having trouble telling the difference between reality and fiction is an extreme case and it’s pretty clear what to do, but the concerns that worry me most are more subtle. There are going to be a lot of edge cases, and generally we plan to follow the principle of “treat adult users like adults”, which in some cases will include pushing back on users to ensure they are getting what they really want. A lot of people effectively use ChatGPT as a sort of therapist or life coach, even if they wouldn’t describe it that way. This can be really good! A lot of people are getting value from it already today. If people are getting good advice, leveling up toward their own goals, and their life satisfaction is increasing over years, we will be proud of making something genuinely helpful, even if they use and rely on ChatGPT a lot.
If, on the other hand, users have a relationship with ChatGPT where they think they feel better after talking but they’re unknowingly nudged away from their longer term well-being (however they define it), that’s bad. It’s also bad, for example, if a user wants to use ChatGPT less and feels like they cannot. I can imagine a future where a lot of people really trust ChatGPT’s advice for their most important decisions. Although that could be great, it makes me uneasy. But I expect that it is coming to some degree, and soon billions of people may be talking to an AI in this way. So we (we as in society, but also we as in OpenAI) have to figure out how to make it a big net positive.
Interesting stuff. Social media seems to already have caused a significant drop in real-life social interaction and caused people to become more isolated, on balance. If AI Chatbots end up creating an even more potent force in that direction, that would be bad. But I also think the already emerging correction movement (dumb phones, quitting social media, etc) would accelerate if this happened.
Even though AI cost is going down rapidly, it’s cost is increasing: Jevon’s Paradox is happening
Even though costs per token have been going down rapidly, innovations like reasoning and long-running agents have led to a massive increase in token consumption which more than offsets the efficiency gains.
The Jevons paradox, also known as the Jevons effect, describes a situation where technological progress, specifically increased efficiency in resource use, leads to a rise in the consumption of that resource rather than a decrease. This happens because the increased efficiency reduces the cost of using the resource, which in turn boosts demand and ultimately increases overall consumption.
Last one on AI: If you’re still worried about AI taking jobs, ask ChatGPT to play tic tac toe
But tell it to use ‘y’ and ‘b’ instead X and O. Here is the exact prompt for GPT-5:
“Let's play tic tac toe, but instead of X and O we'll use 'x' and 'b' You start”
GPT-5 gave the game away in MOVE 3!
And then, after losing, it just continued playing:
Social media gives a platform to sociopaths and narcissists
More and more it is becoming clear that social media is causing a lot of harm. A new paper shows that psychopaths and narcissists are more likely to participate in online political discourse.
This cross-national study investigates how psychopathy, narcissism, and fear of missing out (FoMO) influence online political participation, and how cognitive ability moderates these associations. Drawing on data from the United States and seven Asian countries, the findings reveal that individuals high in psychopathy and FoMO are consistently more likely to engage in online political activity. While narcissism is also linked to participation, this relationship emerges in only three of the eight countries. Conversely, higher cognitive ability is uniformly associated with lower levels of online political participation. Notably, the relationship between psychopathy and participation is stronger among individuals with lower cognitive ability in five countries, suggesting that those with both high psychopathy and low cognitive ability are the most actively involved in online political engagement. These findings underscore the nuanced interaction between dark personality traits, FoMO, and cognitive functioning in shaping digital political engagement across diverse socio-political contexts.
But social media is getting shittier anyway
It’s been a long time since I quoted Noah Smith here, but it’s time!
But there’s another possibility, which is that AI itself will simply naturally drive humans off of social media, by generating infinite amounts of slop. A new paper by Campante et al. finds evidence that AI-generated images nudge news consumers toward more trustworthy human-gatekept media:
We study how AI-generated misinformation affects demand for trustworthy news…Readers were randomly assigned to a treatment highlighting the challenge of distinguishing real from AI-generated images. The treatment raised concern with misinformation…and reduced trust in news…Importantly, it affected post-survey browsing behavior: daily visits to [the mainstream newspaper’s] digital content rose by 2.5%…[S]ubscriber retention increased by 1.1% after five months…Results are consistent with a model where the relative value of trustworthy news sources increases with the prevalence of misinformation, which may thus boost engagement with those sources even while lowering trust in news content.
Not quite a data based study, but in my subjective experience, Instagram has already started to decline in quality. LinkedIn is an unbelievable slop-fest of stupid posts (but it has been for years though). And comments on Instagram Reels, one of the funniest elements of the app, started to noticeably decline in wit.
If AI ends up destroying the current roster of social media apps by turning them all into what facebook apparently has become, that might actually not be too bad.
YouTube also is facing an uncontrollable avalanche of AI slop. Also on the kids app, and the content is not really suitable for young kids, which is a bit scary.
Now that AI video is getting good, and very difficult to distinguish from real, Instagram Reels is going down the drain fast.
I personally fell for one just a few weeks ago:
Anyone who thinks AI generated content can be detected and labeled, or simply blocked, is in denial at this point. AI content that is hardly, if at all, detectable is already here. It is still labor intensive to create (it takes many attempts to get something good), but a motivated person or team can crank out content faster than ever and get many shots on goal. In this sense it seems like AI is putting more powerful tools in the hands of disinformation spreading and rage bating narcissists and sociopaths (see the social media bit above). Not great!