Superior reasoning skills

GPT-4 is supposed to have far superior reasoning skills compared to gpt-3.
However: I rarely see examples demonstrating this.
Does anyone have examples where the reasoning skills of gpt-4 really shine ?

1 Like

Yes … I’ll be posting a bit on wordle in a little while. Search for it :slight_smile:

1 Like

This paper from Microsoft gives tons of examples. Really cool: https://arxiv.org/pdf/2303.12712.pdf

1 Like

In gpt-3, I remember that I asked him to generate a python code for a telegram bot that could talk normally, just like he does, he even made an effort, but he always said that he should have a language model trained to do that being that the openai API already does this, but even so he never gave me a clear answer on how to do this, then gpt-4 arrived, I made the same request to him, initially he did the same thing, I said that the bot was giving nonsensical answers and that he should be able to have a natural conversation with the user, so he rewrote the code, and added a simple prompt that said "you are a smart and humorous bot, answer the following question: " from then it started responding normally, I obviously don’t understand anything about programming, and the fact that GPT-4 did that, albeit much slower, left me stunned, so I think the big difference from 3 to 4 is the ability to know how to interpret and find more creative solutions.

1 Like