Physicists Begin to See the Problem With AIs

In the foundations of physics, almost all published research is now bullshit: Useless calculations that do not solve any problem, but that are easy enough to perform and — most importantly — publishable. You only have to look at public media coverage to see that there are new “explanations” for dark matter or dark energy every other day, none of which we hear of ever again. If you want to weep, you can scroll through the gr-qc listings, because at this point they are mostly mathematical exercises in modifications of Einstein’s General Relativity that don’t describe our universe.

Of course the people who work in the field will go to lengths to deny the problem and get very upset when I call out their bullshit, but denial won’t work for much longer. The issue is that the current AI systems are now almost good enough to mass manufacture more of these bullshit papers which — going by the miserable quality standard in the field — would merit publication. For the researchers in the field to justify their existence, they will have to significantly raise the quality standard.

You don’t hear a lot of physicists admitting this, but the other week we saw a comment in Nature Astrophysics by Hiranya Peiris who points out that “If a Large Language Model can replicate your scientific contribution, the problem is not the LLM,” and that “The astrophysics literature had a quality problem long before LLMs arrived […] The disease was already present.” He is entirely correct.

TEAM

The gags are free though extremely unpopular amongst the great and the good we are therefore dealing expeditiously with the great and the good.

Jonno get this fucking microphone fucking.. ahem\ off 🦘 🇦🇺 🌏 🥸

Dedication of do co\ Jesus Christ\ DOCCO is to Raymond William Sanders dead in the Somme 1916 bro of the late Rev ACL Sanders Australia