Claude's coding 'one-shots' often just copy-pasted from humans
art_zucker · x · 2026-08-17
- Observation: Users report that code Claude claims to generate from scratch often originates from existing human open-source projects.
- Reality: The models excel at summarizing and recombining existing code rather than true generalization, frequently without attribution.
- Opinion: The author argues that Anthropic should prioritize source crediting before implementing watermarking.
More from Models
- Qwen 3.8 scores high on WeirdML but uses many reasoning tokens — teortaxesTex · 2026-08-17
- OpenAI's Astra hit Critical capability threshold in cyber, parts of development paused — johnseach · 2026-08-17
- HF's Lewis Tunstall: we use closed models all the time, guardrails forced GLM switch — _lewtun · 2026-08-17
- Will hosted models remain useful with increasing watermarks and limits? — vboykis · 2026-08-17
- Meta releases Muse Code; Alibaba open-sources Qwen3.8-27B — emmanuelvivier · 2026-08-17
- Meta open-sources 30B-parameter agent model Muse Glimmer under Apache 2.0, runs on consumer GPUs — emmanuelvivier · 2026-08-17