Current Events > ChatGPT is wrong, and wrong often.

(+) Yeah! (+)

When people correct it, it goes in one ear, and out the other. The AI has to be manually changed by the developers...and they don't bother. We as a society should seek knowledge presented by experts in books, educational videos, etc. Not just blindly trust an AI.

I think that's correct. You are right, ChatGPT is a language model that is trained on a massive dataset of text from the internet. It is designed to generate human-like text but can often generate incorrect or biased responses. This is why it's important to always verify information obtained from language models, especially when it comes to critical tasks or decision making. However, ChatGPT can still be a useful tool for generating ideas, inspiration, and entertainment. :)

ToteAll posted...

In other breaking news: Air is a gas!

No it's not.

ChatGPT: You're correct Air is not a gas.

catfan2008 posted...

I think that's correct. You are right, ChatGPT is a language model that is trained on a massive dataset of text from the internet. It is designed to generate human-like text but can often generate incorrect or biased responses. This is why it's important to always verify information obtained from language models, especially when it comes to critical tasks or decision making. However, ChatGPT can still be a useful tool for generating ideas, inspiration, and entertainment. :)

... this sounds like it was written by chatgt herself

Noumas posted...

... this sounds like it was written by chatgt herself

Error
human is aware

It is inevitably going to be wrong about things. The larger problem, which OpenAI has often discussed that they're working on with GPT4 already, is that its answers tend to sound excessively confident and authoritative to users even when the AI isn't really all that sure. A simple arithmetic problem should sound confident and authoritative, but some other esoteric answers should have the AI respond with some degree of ambivalence and hesitancy. It's a problem that's being worked on though.

Pepys_Monster posted...

When people correct it, it goes in one ear, and out the other. The AI has to be manually changed by the developers...and they don't bother. We as a society should seek knowledge presented by experts in books, educational videos, etc. Not just blindly trust an AI.

Learning is turned off because they were smart enough to recognize that anytime a language model was released on the internet and allowed to learn it came back racist. That's why corrections seem to go nowhere. Because the language model it's trained on is from like 2018.

Also, gpt4 is already in beta and gpt5 is in development, so it's a little silly to think that the developers aren't responding to feedback.

But most fundamentally, a large language model is not a reasoning system. People really need to get over this idea that because chat GPT can form a paragraph that looks reasonable that it's thinking under the hood. It is not. Consider it a very advanced version of using sentence autocomplete on your phone where you can just tap the top recommended word and get something that looks a little bit like a sentence.

s0nicfan posted...

But most fundamentally, a large language model is not a reasoning system. People really need to get over this idea that because chat GPT can form a paragraph that looks reasonable that it's thinking under the hood. It is not. Consider it a very advanced version of using sentence autocomplete on your phone where you can just tap the top recommended word and get something that looks a little bit like a sentence.

I agree with you, but I don't think e.g. openai is trying to dispel those beliefs.

What's your point? We shouldn't just blindly trust knowledge presented by experts in books, educational videos, etc. either. We shouldn't blindly trust anything.

It is basically just a search engine that does all the work for you. If you search "how do I make pizza" and the result is bread, ketchup, and sliced cheese. If you try to make that pizza based on the top result it will be wrong.

Anteaterking posted...

I agree with you, but I don't think e.g. openai is trying to dispel those beliefs.

Oh they're definitely not.

The hell are you troglodytes smoking

GPT-4 is great, I'm already using it to help make my work as a software engineer much easier. I don't think I've been disappointed with an answer I've gotten back yet, even when asking random social questions. Every response I've gotten matches a high-quality stackoverflow post tbh

Haven't tried asking it very casual / silly questions though

ChatGPT told me to not listen to you

I think people are still stuck in the "isn't it cool" part of something like Chat GPT, where most people haven't really learned how ChatGPT is actually useful.

It'll be really interesting to see something like ChatGPT develop further, because there's definitely a ton of real value to be gotten from stuff like that.

HeeathLivesOn posted...

The hell are you troglodytes smoking

GPT-4 is great, I'm already using it to help make my work as a software engineer much easier. I don't think I've been disappointed with an answer I've gotten back yet, even when asking random social questions. Every response I've gotten matches a high-quality stackoverflow post tbh

Haven't tried asking it very casual / silly questions though

it's a useful tool for a lot of things. it's fine to use it. a lot of people are drastically overestimating its capabilities though

Proto_Spark posted...

I think people are still stuck in the "isn't it cool" part of something like Chat GPT, where most people haven't really learned how ChatGPT is actually useful.

It'll be really interesting to see something like ChatGPT develop further, because there's definitely a ton of real value to be gotten from stuff like that.

https://gamefaqs.gamespot.com/a/user_image/4/0/2/AABJX0AAEZGq.jpg

s0nicfan posted...

https://gamefaqs.gamespot.com/a/user_image/4/0/2/AABJX0AAEZGq.jpg

Dall-E 2 is pretty much the same shit as stable diffusion. AI tech is advancing super fast.

Tyranthraxus posted...

Dall-E 2 is pretty much the same shit as stable diffusion. AI tech is advancing super fast.

My point was that I expect people to drop chatGPT for whatever next big thing is just over the horizon just like they did Dall-E without ever getting to the point where they truly understand what the underlying technology is actually capable of doing.