I keep seeing this boosted https://mastodon.social/@Lana@beige.party/116195578176984752 and I found the "recent research study" (which isn't linked) and it's from 3 years ago and about Claude 1 and ChatGPT 3.5. Here it is: https://arxiv.org/html/2305.04388v2
Spoiler: The "LLMs do not think" wasn't what the research conclussion was either.
The whole point of the paper was about bias in LLMs which points out racial bias in early models which patently bad. Thankfuly, three years later but Mastodon LLM claims are still pure ๐ฉ ๐คท