https://chat.openai.com/share/f7ad44ba-f00c-41b8-b0b1-8763795724a5

This is probably the most serious phail I've seen from gpt4 in a while.

That said, it's still impressive! Only one year ago these kinds of phails were the norm; the fact that I'm now surprised by it is a sign of progress!

Previous research shows that GPT-4 has shown a decline in intelligence. Its success rate in solving prime number problems dropped from 97.6% to 2.4% between March and June. GPT may still exhibit significant randomness and unpredictability in mathematical reasoning.

Previous research shows that GPT-4 has shown a decline in intelligence. Its success rate in solving prime number problems dropped from 97.6% to 2.4% between March and June. GPT may still exhibit significant randomness and unpredictability in mathematical reasoning.

https://twitter.com/svpino/status/1681614284613099520

crypto-anarchist · onchain social explorer, advisor · spread FC gospel in /chinese community · curate SocialSight.telegram· write 0xluo.xyz · host /luo ·中文 @luo

GPT-4 is very explicitly not good at math by design. This is why the Wolfram plugin exists. Always always activate the Wolfram plugin if you care about doing math.

Judging GPT-4 for how it does math is like judging a lawn mower for how well it paints landscapes.