Researcher flags severe LLM degradation spreading across sessions on the same machine
doodlestein · x · 2026-09-20
Researcher doodlestein discovered that the severe LLM behavior degradation he reported earlier also appeared in other sessions on the same machine, joking that "a ton of people got a massive dose of SHAME."
In the original post, he argued this behavior is critically important to understand if LLMs are to be used in mission-critical settings, noting he had never seen that level of degradation in Claude or ChatGPT—suggesting a model-specific systemic failure.
Related event: Gemini Flash 3.8 Bizarre Breakdowns; Google Blames Hardware Service Failure(9 posts)→
More from Models
- a16z's Martin Casado: Google's Carbon is 'really good at code' — xennygrimmato_ · 2026-11-12
- Google Cloud Now Hosts Gemini, Claude, and GPT — a Hyperscaler First — samcharrington · 2026-11-12
- Carbon + /swarm in Antigravity: Big Model Thinks, Lighter Models Work — samcharrington · 2026-11-12
- Gemini 4 Drops Pro/Flash for Element Names: Argon and Carbon, Carbon Claims 2x Cheaper Per Task — samcharrington · 2026-11-12
- Written Off Weeks Ago, Google Ships Dozens of Gemini and Cloud AI Updates — samcharrington · 2026-11-12
- The Zvi opens reaction thread on Gemini 4 Carbon: is it worth extended coverage? — TheZvi · 2026-11-12