Jackietrader
Jackietrader

Asking Humans vs GPT to choose a number between 1-10

Very interesting that GPT3.5 has similar characteristics to human respondents.

Post image
16mo ago
salt
salt
Gojek16mo

Perhaps it can be deduced that corpus of data accumulated on the internet 🛜 contains subliminal information ℹ️ about human behaviour.

Jackietrader
Jackietrader
Google16mo

Could very well be. @salt It is becoming obvious that Large Language Models have moderate generalizability. Likely, due to the increase in number of parameters that the LLMs are being trained on to create foundation models.

salt
salt
Gojek16mo

@Jackietrader Ofcourse, however in my opinion we are at the exact inflection point where this technology just goes hypersonic.

Currently GPT3/3.5/4 based models do good enough that they can beat the bottom quartile.

It would be interesting to see, how good they can get now.

Jackietrader
Jackietrader
Google16mo

It’s kinda insane how LLMs are mimicking human behaviour in the unlikeliest of places.

salt
salt
Gojek16mo

Very interesting 🤔

Pixeldynamic
Pixeldynamic

Is it? GPT 3.5 looks like skewed Gaussian curve whereas human behavior looks like an S curve with an anomaly at 7

salt
salt
Gojek16mo

Yeah true. But E(X) = 7 in both cases.

Droughtbear
Droughtbear
Student16mo

Pick random number from 1-10….. 47 humans pick zero. Chad move.

Jackietrader
Jackietrader
Google16mo

😂😂😂😂

Humans just too 🤪

Discover more
Curated from across