Hacker News Reader

Comment by Xcelerate | Hacker News Reader

It seems quite clear that inferring traits about people is something LLMs will only become better at with time.

Everyone claims MBTI is akin to astrology, which means there should be no predictive capacity of those four letters. But just for fun, I gave GPT-4 the top 20 songs of my music playlist and asked it to guess my MBTI type, and it correctly did so. I repeated the process with two friends (asking their MBTI type prior to querying GPT) and it guessed them correctly as well. If MBTI is no different than random noise and GPT lacks inference capabilities with regard to humans, the probability of this occurring due to chance alone is less than 1 in 16^3 (*approximately—non-uniform distribution of types).

View on HN

I don’t think anyone says MBTI is random. I think the doubt is in the predictive power of a whichever MBTI classification. The test for your MB label is pretty well designed to extract “some” information. Academics just don’t think the labeling is particularly valuable.

Fair point. But are the “superior” systems like Big 5 (or more recent approaches that utilize independent component analysis) really that much more predictive than MBTI? It seems like there is a lot of squabbling over the difference of a few percentage points in information gain.

To a layman, they're all useless, and in a dangerous way: they transport a sense of understanding a person/character/psyche that they don't support.

A professional can gain _some_ information from all of them. Just that the Big 5 have much more literature behind them than MBTI, so there's more reference data for any given constellation. That's a reinforcement loop. Methodical differences and reliability and all that fluff can be used to argue in more detail in favor of one versus the other, but by now, it comes down to what folks are more familiar with.

In the end, these are just a few out of many, many tools of the field. If you see somebody build their career on "(MBTI|Big 5|...) analysis" and nothing else, they're not a professional but a quack.

Everyone knows you can fully categorize humans using only two systems: Blood Type and Hogwarts House. Birth month is hogwash, of course.

Just for fun I tested this as well and while GPT-3.5 was wrong, GPT-4 was correct

Reply on news.ycombinator.com