Claim: Anthropic Broke Opus with Text Watermarks
nikvassev · x · 2026-08-21
A user claims that Anthropic likely degraded the Opus model by shoving text watermarks into it, resulting in hallucinations and mangled sentences. The criticism contrasts the model's current performance with the company's denial of any impact.
More from Companies & People
- Dean Ball jokes he's an OpenAI culture fit: he runs three overlapping blogs — deanwball · 2026-08-21
- Data quality is the #1 barrier to enterprise AI, accounting for 70% of success — kashifmanzoor · 2026-08-21
- Poolside acquisition highlights a new exit path for open-source AI companies — sachdh · 2026-08-21
- Brain Drain to 'Silicon Tower' Sparks Concern Over Independent AI Researchers — TuhinChakr · 2026-08-21
- Open Weights Models Face UX Friction; Both Closed and Open Will Grow — JosephJacks_ · 2026-08-21
- VC on DeepSeek's open-sourced harness: generic agent layers will commoditize — mattsheehan88 · 2026-08-21