ChatGPT might not be coming for all of us quite yet...

quote:

---

He was saying how literally all it does is scour the internet for information and present it. It doesn't care if the information is right or wrong, it just copies it and pastes it in whatever order it determines. He said the reason people say "Thaat sounds like it was written by a human" is it literally was written by someone before. It's basically one huge plagiarizer.

---

quote:

---

quote:
He was saying how literally all it does is scour the internet for information and present it.

That’s a pretty massive oversimplification. It’s not actively scouring the internet. Its “knowledge” is limited to the dataset that was used for training (which is, admittedly, massive).

That being said, the conversation around where GPT gets its “information” largely misses the point. Large language models (LLMs) like GPT are designed to learn language, first and foremost, by consuming human written language and developing an understanding of relationships between words. It’s not that much different than how an infant learn language - it builds associations, tries them out, and is then corrected (via training) when it’s wrong.

I bring this up because people often talk about GPT in terms of a “search engine,” saying it doesn’t actually understand the meaning of its words. Creating better search engines is certainly one application for LLMs but that’s not why they are important.

LLMs aren’t currently expected to be artificial general intelligence. They are just a piece of AGI. It would be like a human who could read and write words on a page but had no other senses. No vision beyond the page, no hearing, no ability to touch/manipulate objects. The human would only “understand” the words on the page as.. well.. words.

But the real jump in AI happens when you combine these LLMs with other AI research. You merge GPT with an image recognition AI and a camera. Now it has sight and can associate words with images. You merge it with a speech recognition program. Now it can have a conversation with a human.

And then you get to the scary part - you give it access to the internet. Now it can actively search out knowledge it doesn’t already possess AND potentially interact with the outside world in a more uncontrolled environment. That means it can take actions much more freely and see the results if those actions. The logical next step is to give it a directive and see what it does. Maybe it cures cancer, maybe it kills us all.

The thing is, there’s an argument that the language part is the most complex and difficult to achieve. That’s why people think GPT and other LLMs are such a big deal.

---

quote:

---

It doesn't care if the information is right or wrong, it just copies it and pastes it in whatever order it determines. He said the reason people say "Thaat sounds like it was written by a human" is it literally was written by someone before. It's basically one huge plagiarizer.

---