Hacker News Reader

Comment by code_runner | Hacker News Reader

These models aren't capable of "doing" things. They can't necessarily calculate values etc. They can assemble text based (more or less) on statistical analysis of all their input data, and the are incredible at it... but if you ask what 2+2 is, its not doing math to get to 4.

They certainly can't deploy the code they generate etc.

View on HN

Things are moving so damn fast, while you were correct when you posted - two hours later you were not:

https://twitter.com/gdb/status/1638971232443076609

I tried asking GPT-4 a math question and it does do math:

Me: "What (8 * 25) / 14 + 4 - 2?"

AI: "Let's break down the expression and solve it step by step:

    Multiply 8 and 25: 8 * 25 = 200
    Divide the result by 14: 200 / 14 ≈ 14.2857 (rounded to four decimal places)
    Add 4 to the result: 14.2857 + 4 = 18.2857
    Subtract 2 from the result: 18.2857 - 2 = 16.2857 (rounded to four decimal places)

So, the result of the expression (8 * 25) / 14 + 4 - 2 is approximately 16.2857."

>its not doing math to get to 4

OpenAI just announced a partnership with Wolfram Alpha, so now it can ask a different computer to do the math.

It's only time till they have access to your filesystem and then that's it

Reply on news.ycombinator.com