Researchers Find Abnormal Tokens Causing Weird DeepSeek-V3 Outputs

Researchers found anomalous tokens in DeepSeek-V3 and r1 that trigger bizarre behavior, similar to earlier GPT-2/3 findings, with the issue tied to training-time tokenization pathologies.

2026-08-21 ~ 2026-08-21 · 2 related posts

• DeepSeek-V3 and r1 Found to Contain Anomalous Tokens Causing Bizarre Behavior — teortaxesTex · 2026-08-21
• Dirty Tokens Cause Model Anomalies Linked to Training Pathologies — teortaxesTex · 2026-08-21