ChatGPT Fails And Limitations | Artificial Stupidity
Vložit
- čas přidán 21. 07. 2024
- In this video, I'll go through some limitations and fails of the current generation of chat AI models, focusing on OpenAI ChatGPT, and of course GPT 3 behind it. I'll talk about the general limitations, show a lot of failures and mistakes in the answers that are already happening, and share some of my thoughts on how they can and should not be used. Also some ideas on if it will make some jobs extinct or not.
As always, if you appreciate the content here, show it by clicking the thumbs-up button, and use the comments section for feedback or future video requests as you like. And if you subscribe to my channel, you will better notice when there's new content on my channel.
Here are some fun background links:
- • ChatGPT writes my code...
- github.com/giuven95/chatgpt-f...
- / chatgptcoding
- / promptcoding
Good to see a video on chatgpt that is not sensationalist nonsense for once. Well done. Sir 👏
Thank you!
Okay, here's one to wrap our minds around, experts are trying to say these systems aren't as intelligent as we think, while simultaneously loading something the experts now claim is dumb into our defense systems and everything, WTF? Why would we want something so stupid running all our systems, it either is intelligence or it isn't and either way it doesn't sound good, it does not compute, pun intended.
Haha, that's true. My thoughts: It's a tool, and right now many are trying to bend it to do things that are not a good fit. So there are valid cases, and then there are a lot of horribly bad cases to use modern LLMs.
But the valid cases are interesting, yes!
@@DevXplaining a lot of tools have been used to bludgeon people to death, good cases just don't seem to be outweighing the bad cases on this "tool." Perhaps it is us being the tools trying to even find a place where this technology should applied, I don't think the world is ready for AI by any means. I feel this is what Hawking was warning us about.
Good points. They fit with my experience using ChatGPT as well. I disagree with a couple points though. On a fundamental level, ChatGPT's linguistic intelligence isn't that much different from humans'. Both intelligences are algorithms based on neural networks. The differences are in how & when they are trained and on what body of information. In humans, this training is a more wholistic process. Also, ChatGPT indeed does know that 2+2=4 and you'll have a hard time convincing it otherwise (it's probably possible, but it's also possible to convince humans as well) IF this fact is the primary topic of discussion. However, if it's a side calculation, it can easily become confused and give you the wrong numbers or solve the wrong problem. It has a limited number of things it can focus on in one response. It does better with small, bite-size tasks rather than longer processes, as it doesn't have an infinite "attention span", to use a human analogy.
Hi, thank you for your comment, much appreciated!
I somewhat disagree on the capabilities, but would probably require an actual discussion to see if we are meaning the same things - might be we actually agree. I just don't really see these as same, although in a wide enough view, I admit yes they are.
On 2nd part, yeah, I haven't seen that arithmetic mistake myself, I saw some screen grabs but half of them are memes these days anyway. And yes, it's not any different than persuading humans, I agree. Furthermore, GPT4 was just released, and while I'm just getting to understand the changes, it seems more capable already, it's a refined experience. Need to look more at it. They still have their fails, but haha, have to say the speed of evolution on this field right now is mindblowing.
That is not how brains work, like at all
I agree with you .. i feel we are in the same path.. i am writing a review on ChatGTP .. and if you agree I will use your content and cite you ,, is your name Arto Santala? thank you bro
Hi, thank you for feedback! Well it's publically available content and mostly opinions and viewpoints, so if you wish to refer to it, go ahead.
Worth mentioning that while the big guidelines still are valid, GPT-4 was published week after I recorded this, and does change a few things, like opens up token limits and enhances capabilities. AI is catching up fast :)
As an large language model I can only output mathematically plausible but entirely bullshit answers to anything requiring the tiniest amount of complexity and technical accuracy.
Haha, true, that's happening even more and more these days. I got that today even for a coding question, askin ChatGPT to optimize something for efficiency. :)
Oh boy; nobody knows what is right and wrong. There is not one hing all humans consider wrong or right.
Haha, true that :)
1+1=2
plz chatgpt isn't even emulation yet. At most it's just siri or alexa level bot. We need to code another half of the brain to use ChatGPT properly as emulation.
Heh, true. We'll see interesting developments within this year and the next one, for sure :)