GLM-5.3 abliterated for a month, no takeover: dev rails against AI doom narratives
mariofilhoml · x · 2026-10-13
A developer cites evals showing GLM-5.2-and-stronger models can already self-replicate end to end — with small models projected to match that by year's end — to mock AI doomers: the "dangerous" abliterated GLM-5.3 has been out for over a month with no takeover. A snapshot of the ongoing safety-eval vs. doom-narrative fight.
More from AGI Musings
- Ethan Mollick: OpenAI and Anthropic compete at the cheap end, not just the frontier — eldonredwards · 2026-10-13
- AI consultant is now the second-fastest growing US job, per LinkedIn — vasuman · 2026-10-13
- The UI Era Is Ending: Otto's Mastra Talk Argues AI Is Making Interfaces Disappear — DynamicWebPaige · 2026-10-13
- Personal agents will eliminate friction: recouping subscriptions, refunds and tax appeals — mon__lim · 2026-10-13
- Machine consciousness draws the fiercest pushback of any AI debate, says dev — abhiadesai · 2026-10-13
- Chollet: agriculture is 'self-replicating, recursively self-improving open-source nanotech' — fchollet · 2026-10-13