Google dusts off the failed Google+ playbook to fight ChatGPT

Re:Google dusts off the failed Google+ playbook to fight Cha... by jvanderbot on Friday March 10, 2023 @07:42AM (#35093655)

Well, you can't train LLM on "facts alone", since they are basically throwing darts at the next word, not using information to craft a message. But I hear your larger point: it'd be nice to have an agent that was trained on wikipedia information, but already spoke a language

[ Parent | on HN ]

Re:Google dusts off the failed Google+ playbook to fight Cha... by r3trohack3r on Friday March 10, 2023 @09:16AM (#35094520)

GPT is trained on Wikipedia though.

It’s speculated libgen is in there as well.

[ Parent | on HN ]

Re:Google dusts off the failed Google+ playbook to fight Cha... by artificial on Friday March 10, 2023 @09:46AM (#35094886)

Let’s not forget Twitter was a training source. Early Twitter was like 4chan.

[ Parent | on HN ]

Re:Google dusts off the failed Google+ playbook to fight Cha... by jvanderbot on Friday March 10, 2023 @10:34AM (#35095666)

I think GGP comment started this thread about a knowledge engine, not a "word generator" if that makes sense. Training on Wikipedia in this case means "training to put together words like wikipedia authors do" right? Not "learning the facts of wikipedia to provide useful synthesis and cross referencing." though it probably does OK by accident.

[ Parent | on HN ]