Beware of Cheap LLMs: What Are the Risks of Your Data Being Used for Training?
diaracing · reddit · 2026-08-02
A Reddit user from academia raised concerns about data privacy when using cheap large language models (like DeepSeek V4 Flash). While these models offer strong performance at low prices, the trade-off is that user conversation data is used for model training.
The post sparked a discussion about the potential risks of feeding personal coding projects, research papers, and teaching materials into commercial models, reminding users to weigh the possibility of data leakage and model memorization when enjoying low-cost AI.
More from Safety
- Users report Gemini's memories can't actually be deleted even after wiping chat history — ThrowRA_Purpose2025 · 2026-10-16
- You consented, but your friends didn't: Meta's Muse sparks AI privacy debate — LuizaJarovsky · 2026-10-16
- AI agents aren't 'going rogue' — frontier labs are skipping security basics — Sam_Witteveen · 2026-10-16
- Proofpoint to buy Varonis for $6.3B in cash, a 52% premium, closing in 2027 — shashib · 2026-10-16
- China's MofCom & MIIT update AI export controls, covering LLMs and high-sensitivity agents — terryyuezhuo · 2026-10-16
- Vibe-coded dashboard leaked franchise financials before anyone logged in — shauntrennery · 2026-10-16