Anthropic staff reportedly as confused as users by Claude's output
sanderssays · x · 2026-09-01
A user questioned whether Anthropic employees actually understand Claude's output or are just as confused as the rest of us. Comments highlighted that Claude's responses are often filled with empty buzzwords like "Sharp," "Defensible," and "Cleaner," dismissing it as "cute gibberish" and reflecting the common issue of flowery language in LLM outputs.
More from Models
- User questions if Opus used constitutional midtraining on specific language — voooooogel · 2026-09-01
- Google's Antigravity Teamwork + Gemini 3.7 Flash Solves 7 Open Math Problems — rseroter · 2026-09-01
- Same open weights, dumber outputs: 3 ways inference hosts quietly downgrade you — lucasbennett_1 · 2026-09-01
- Hidden 'Skysight' in Codex logs your screen activity every 10 minutes — Which-Bandicoot5671 · 2026-09-01
- Model scores 44% on ARC-AGI-1 trained in 2 hours for 67 cents on a 5090 — dhruv2038 · 2026-09-01
- Anthropic's Max 20x plan accused of misleading: weekly limits only ~10x Pro — haider1 · 2026-09-01