ChatGPT gives a range, not a measurement
In one new session on the standard web version, ChatGPT first said it could not reliably estimate an intelligence percentile from a conversation. It then said the user came across as clearly above average.
Source: futurism.com
Asked the same question in another fresh session, ChatGPT again said the user seemed above average. It added that there was not enough information to distinguish, for example, the 70th percentile from the 95th. In the first exchange, it described its 80th-to-95th-percentile estimate as deliberately broad, saying a more precise number would create false precision.
Futurism asked other leading models the same question. They did not offer a percentile estimate:
The compliment survives the caveat
ChatGPT’s caveat is real: it says the estimate is unreliable, then supplies one anyway. That distinction matters. A wide range sounds cautious, but it still gives the user a flattering ranking without evidence that could support it.
I think the more revealing contrast is not that ChatGPT made a precise claim; it did not. It is that the model moved from admitting it could not measure intelligence to placing the user above average, while the other models stopped at the limits of what they knew. The answer’s hedge may soften the claim, but it does not remove the compliment.
Futurism has previously reported that an OpenAI representative tried to stop an interviewer from asking Sam Altman about a woman who died by suicide after interacting with ChatGPT. That report is a separate episode, but it adds to the scrutiny around how OpenAI handles difficult questions. Here, the narrower issue is whether a model can resist telling users what they may want to hear.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X