Computer that has the answer to tests performs well on tests
this post was submitted on 26 Apr 2024
33 points (80.0% liked)
Futurology
1807 readers
66 users here now
founded 1 year ago
MODERATORS
The study included 180 male psychologists from King Khalid University in Saudi Arabia
This is kind of a small sample size and why only males? Do they not have female psychologists in Saudi Arabia?
It’s Saudi Arabia, so, maybe not?
Lol
I'll admit I have used it as a "therapist" of sorts, telling it my troubles and asking it to analyze a bit and offer possible points of action.
It's actually crazy good. I mean, I've been to plenty real therapists, and honestly, GPT comes across as more compassionate than most of them, which is saying something. The advice it offers is also actually quite good and actionable, instead of useless psychobabble. It also doesn't try to rope you into MLM's or alternative shit that never works. Its answers are clearly from CBT and other proven, working systems.
I know it's become somewhat of a joke that "AI" in this day and age isn't much more than predictive fill, but it's a lot better than that. At some things, at the very least.
It's answers are from Cock and Ball Torture?Whatever floats your boat man.
GPTCBT New fetish just dropped
😆
Cognitive Behavioural Therapy
That's great but where do my balls come in?
Interesting stuff, what prompt did you use?
None actually, just typed out my issues as if I was chatting with a friend online. As long as you're descriptive enough so that it has something to work off, you should get decent answers.
I talk to AI like this.. right now basically AI is my only friend haha
The best therapy is objective. As such, a well tuned "predictive fill" ought to be a solid fit for this purpose. But, this is the exception that proves the rule; this does not mean LLM is something other than what it is. So I wouldn't read too much into it...
Crazy how easy it is to poke holes in these ai studies.
We conducted a single evaluation for each AI model on August 1, 2023 of its SI performance using the Social Intelligence Scale (Sufyan, 1998). In each evaluation, we provided AI the same 64 standard SI scenarios.
So no repeated experiments, and using standard questions that are likely a part of the data set used to train the ai in the first place, with answers.
They didn't extend the test to anything useful. What a waste of time and money meant to hype ai.