News for hackers. Stuff that matters.
Re:GPT-4 Can't Reason by Rapzid on Tuesday August 08, 2023 @12:35PM (#37051676)
Couldn't she have been dead at noon in the colloquial sense and then resuscitated at 12:01? I can't tell if that defense is being pedantic, or criticizing its answer is being pedantic.

Maybe both.

Also, I hate how GPT always offers its unsolicited opinions and advice when answering questions:

> Remember, medical information can be complex and nuanced. Understanding a person’s health status often requires a comprehensive examination and should be performed by a healthcare professional.

Re:GPT-4 Can't Reason by sheepscreek on Tuesday August 08, 2023 @12:40PM (#37051755)
My first thought was - what you mean by alive? A person can be brain dead. And alive (with vitals) at the same time. With some assumptions, one can say she was alive at noon.

However, I believe OpenAI has added significant safeguards to prevent any such assumption, to the point where GPT-4 isn’t very useful at times.

This may be fixable with prompt engineering if it can be persuaded to let it’s guard down a bit.

Re:GPT-4 Can't Reason by falcrist on Tuesday August 08, 2023 @12:38PM (#37051720)
That's not unsolicited advice from GPT so much as it's a mandated CYA disclaimer from OpenAI.
Re:GPT-4 Can't Reason by Rapzid on Tuesday August 08, 2023 @12:39PM (#37051744)
Perhaps in this case of medical questions, but GitHub Copilot Chat will start ai-splaining code readability when you ask it questions about concise syntax so.. It's a total mansplainer.
Re:GPT-4 Can't Reason by benjaminl on Tuesday August 08, 2023 @12:43PM (#37051809)
GPT-4 gave me a concise correct answer without disclaimers.

This may be due to using many custom instructions like the following in ChatGPT settings:

- No need to provide disclaimers about your knowledge cutoff

- No need to mention you are an AI language model

- Only discuss safety when it is not obvious and very important

- You should act as an expert in the relevant fields

Re:GPT-4 Can't Reason by Terretta on Tuesday August 08, 2023 @12:53PM (#37052007)
These are excellent, although for instructions I've had issues with compound negatives (e.g. "no need") and better luck with single word affirmative imperatives (e.g. "avoid").
Re:GPT-4 Can't Reason by quijoteuniv on Tuesday August 08, 2023 @12:53PM (#37052000)
I got used to not reading the last paragraph of the answer already