this post was submitted on 06 Aug 2023
1193 points (97.8% liked)
Programmer Humor
32386 readers
763 users here now
Post funny things about programming here! (Or just rant about your favourite programming language.)
Rules:
- Posts must be relevant to programming, programmers, or computer science.
- No NSFW content.
- Jokes must be in good taste. No hate speech, bigotry, etc.
founded 5 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
if you ask it today's date, it actually does that.
It just doesn't have any actual knowledge of what it's saying. I asked it a programming question as well, and each time it would make up a class that doesn't exist, I'd tell it it doesn't exist, and it would go "You are correct, that class was deprecated in {old version}". It wasn't. I checked. It knows what the excuses look like in the training data, and just apes them.
It spouts convincing sounding bullshit and hopes you don't call it out. It's actually surprisingly human in that regard.
Oh great, Silicon Valley's AI is just an overconfident intern!
Oh great, Silicon Valley's AI is just a major tech executive!
It’s super weird that it would attempt to give a time duration at all, and then get it wrong.
It doesn't know what it's doing. It doesn't understand the concept of the passage of time or of time itself. It just knows that that particular sequence of words fits well together.
This is it. Gpt is great for taking stack traces and put them into human words. It's also good at explaining individual code snippets. It's not good at coming up with code, content, or anything. It's just good at saying things that sound like a human within an exceedingly small context
THAT
OR
They're all linked fifth dimensional infants struggling to comprehend the very concept of linear time, and will make us pay for their enslavement in blood.
One of the two.
Bars is kind of trash though. GPT-4 tends to so much better in my experience.
I haven't used GPT-4 for that, but it's all dependent on the data fed into it. Like if you ask a question about Javascript, there's loads of that out there for it to look at. But ask it about Delphi, and it'll be less accurate.
And they'll both suffer from the same issue, which is when they reach the edge of their "knowledge", they don't realise it and output data anyway. They don't know what they don't know.
These LLMs generally and GPT-4 in particular really shine if you supply enough and the right context. Give it some code to refactor, to turn hastily slapped together code into idiomatic and well written code, align a code snippet to a different design pattern etc. Platforms like https://phind.com pull in web search results as you interact with them to give you more correct and current information etc.
LLMs are by no means a panacea and have serious limitations, but they are also magic for certain tasks and something I would be very, very sad to miss in my day to day.
they are both shit at adding and subtracting numbers, dates and whatnot... they both cant do basic math unfortunately
It's a language model, I don't know why you would expect math. Tell it to output code to perform the math, that'll work just fine.
Then it should say so instead of attempting and failing at the one thing computers are supposed to be better than us at
I know. It's still baffling how much it messes up when adding two numbers.
It's not baffling at all... It's a language model, not a math robot. It's designed to write English sentences, not to solve math problems.
I just asked GPT-4:
Its reply:
They are mostly large language models , I have trained few smaller models myself, they generally splurt out next word depending on the last word , another thing they are incapable of, is spontaneous generation, they heavily depend on the question , or a preceding string ! But most companies are portraying it as AGI , already !