Doctor Published Research Papers with Breakneck Speed. ChatGPT Wrote Them All | Hacker News Reader

Doctor Published Research Papers with Breakneck Speed. ChatGPT Wrote Them All

thedailybeast.com

33 pointsbookofjoe9 comments

This is pretty much the last thing academics needs, at least with the current system as it is.

On the other hand, there's something fitting about the artifice of it.

How does the AI aspect impact anything? The doctor puts his name and face to the text, his reputation and belief in his merit is on the line, nobody else's.

Hear me out: LLMs should store their responses so that when someone asks them "did you write this?", even with modifications of the response they can confirm or deny. Since they are costly to operate, it will cut down on their abuse.

Assuming this is a serious suggestion and not a joke that I have missed, there are so many cloud LLMs that this kind of search is not feasible. And local LLMs that you can run on your own hardware makes detection completely impossible. It's better to acknowledge that people will use LLMs, and judge the papers based on their content and not their writer.

I was serious, I meant the cloud LLMs, the ones you run on your own are out of scope

This will never work, fundamentally any computer is unable to solve certain types of problems. They run into several different core computer science problems when you try to have them do this.

There is the halting problem and the decidability problems. You can approximate certain things using a perceptron/neuron floating point configuration, but they are only approximates; and do nothing towards addressing either of those problems.

Even with an approximation? "That sounds like something I might say", people are capable of doing this, why can't LLMs?

Spooky action at a distance?

Kidding aside, no one knows what exactly we do differently that allows for complex thought, context switching, or conflict resolution that works around those aforementioned problems.

Reply on news.ycombinator.com