Cisco's Test Exposes DeepSeek's Weaknesses
Cisco put DeepSeek's open-source model, DeepSeek R1, to the test and it failed to handle all 50 harmful prompts from the HarmBench dataset. This makes DeepSeek R1 the least effective model so far, with competitors like Meta's Llama 3.1 and OpenAI's o1 showing much better performance. The results are raising alarms about the model's vulnerability and potential data security risks.