Post

Of course Emily Bender is boosting an article promoting the false dichotomy that either AI can never be used to give feedback that could help improve a scientific paper OR we have to get rid of peer review. How about—get this—we USE MORE THAN ONE METHOD TO GET POTENTIALLY-VALUABLE FEEDBACK.

We don't disregard all human opinions on the technical elements of writing just because word processors have spelling and grammer checking! If you have a flaw in your analysis, do you want more or less tools that might help you find it?

But this makes perfect sense because Emily Bender is not a scientist. Linguistics is a science, yes, but the goal of a scientist is to better understand reality. Bender has repeatedly demonstrated she has no interest in understanding reality.

She is religiously dedicated to maintaining ignorance of any and all LLM capabilities that might contradict statements she has made. Since she has stated in the past that LLMs HAVE to be useless, this means she has shielded herself from any information about LLMs' abilities to do literally anything.

This is not to say that there are not moral, ethical, social, and political problems with AI: there are many. But because she has shielded herself from essentially all information on the subject, Bender is largely incapable of addressing these issues at anything beyond the most basic level.

“This is not to say that there are not moral, ethical, social and political problems with AI” what problems? what do you have in mind when you say that?

We should not be opening up more fossil fuel plants to power datacenters when we as a country urgently need to be expanding our supply of renewable energy, AI or no. There are numerous datacenters that have even more extreme environmental issues. People should be in jail over those.

The presence of AI complicates establishing truth by making it easier to forge evidence.

It's vitally important that a handful of coporations do not de-facto control all use of a new and powerful technology so that power over society is not so concentrated. However, open-weight models, the best counterweight we have, enable other kinds of harms.