Even more convincing when it is wrong?

ChatGPT lacks the ability to prove that output is factually correct (which is an interesting but distinct problem). However, as long as the answer is derivative of supplied information, ChatGPT shines. Even today it can already:

- Write coherent short stories based on a prompt

- Be a personal diary/assistant

- Summarize and extract information from text.

There's a lot of applications for that.

It won't be wrong forever, there's a good chance they'll figure out how to evaluate sources and provide more useful results. And besides, the bar is low -- Google search results for some topics (e.g., programming related) are quite poor, filled with garbage and clone sites.

I think people saying that just don't understand what GPT is. It's not a database of knowledge that sometimes invents stuff. It's a statistical model that ALWAYS invents stuff. It will never be able to evaluate sources, because it doesn't use sources.

isn't "evaluating sources" the exact problem of web search that Google has been throwing billions of dollars at for many years, with an equivalently dedicated and motivated horde of SEO specialists on the other side? How is chatgpt going to easily solve that overnight?