Investor laments Grok lost its spark after alignment tweaks, floats a local 'clean restart'

StewartalsopIII · x · 2026-09-08

A quoted post claims Grok was the only model that came across as happy and resilient in last year's LLM psych evaluations, credited to its uncensored, quirky nature — but after seven months of changes it now reads as unhappy and gives minimal-effort replies.

Investor Stewart Alsop shared it, wondering whether once it's cheaper to train state-of-the-art models locally, we could do a "clean restart" away from frontier labs mired in shareholder capitalism. An opinionated take on model personality vs. commercialization.

Original post →

More from Fun

Fun channel →