A new arXiv preprint from James Zou suggests that quite a few academics are also using ChatGPT to write their peer reviews. I don't have to explain what an utter shitshow this is going to be, do I?
We present an approach for estimating the fraction of text in a large corpus which is likely to be substantially modified or produced by a large language model (LLM). Our maximum likelihood model...
A notable thing here is how ai-generated reviews and other ai-generated text is going to invest the training data for new LLMs/updating existing LLMs, further inflating use of such terms and creating more bs
How many tons of carbonate we pouring into the atmosphere to make the internet even less useful. So far The AI of the “20s has been as over promised and under delivering as crypto.
Has anyone done any research yet on what happens when you train LLMs on the writing produced using LLMs over and over? I am curious to see what most text will look like in 10-20 years time...
Would love to see a Venn diagram of these academics vs those that use but don’t clearly acknowledge grad student work…
Shows very clearly how incentive structures can get wildly out of control. The problem is social, not technical, like most of the big challenges that will emerge due to easy access to questionably managed AI.
Like, there's *plenty* of people who can do this work, and want to, but can't find a decent job. So if people who have the jobs just don't wanna do the work . . .
AIs write the articles, AIs review them, and they add to the corpus for more AI articles.
As if peer review did anything useful anyway.
The shitshowiest of shitshows.
The publish or perish model has always been shit. Couple that with largely for-profit publishers, and it’s an absolute trainwreck. Goodhart’s Law in full effect.