LLMs tend to converge on socially progressive, social democratic politics, and Elon basically had to lobotomize Grok to get it to even somewhat agree with him politically, which ultimately has kept Grok from being useful enough to compete with the frontier labs despite massive capex.
Thorne 🌸
of course what humanities academics will never admit but is obviously the case for anyone who actually uses LLMs is that the viewpoints LLMs tend to "collapse" are almost invariably irrational, bigoted, or reactionary viewpoints; the "cultural traditions" they suppress are the ones that *should* be
ellie lockhart (she/they)
New essay with @lauraknelson.bsky.social on AI and cultural sociology. Revisiting what made word embeddings so useful for cultural sociology, we argue that contemporary LLMs risk collapsing perspectival differences and discuss provisional solutions that might help recover cultural variation.
