DeepSeek, a Chinese AI startup, recently faced scrutiny after its chatbot was evaluated by NewsGuard, revealing a staggering 83% fail rate in providing accurate information about news topics.
This performance ranked DeepSeek’s R1 model 10th out of 11 chatbots tested, significantly trailing behind its Western competitors, which had an average fail rate of 62%.
Key Findings from the Audit
-Accuracy Rate: DeepSeek’s chatbot achieved only 17% accuracy in delivering reliable news and information.
– False Claims: It repeated false claims 30% of the time and provided vague or unhelpful answers 53% of the time.
– Debunking Falsehoods: The chatbot managed to debunk provably false claims only 17% of the time, indicating a severe lack of reliability in handling misinformation.
DeepSeek’s rapid rise to popularity—becoming the most downloaded app shortly after its launch on January 20, 2025—has raised concerns regarding the implications for misinformation and the broader AI landscape. The chatbot’s performance has sparked debates about the reliability of AI technologies that claim to compete with established models like OpenAI’s ChatGPT, especially given that DeepSeek reportedly trained its model at a fraction of the cost (approximately $5.6 million) compared to its rivals.
Moreover, the audit revealed that DeepSeek often incorporated the Chinese government’s stance in its responses, even when questions were unrelated to China. This pattern raises alarms about potential biases and the manipulation of information by state narratives.
While DeepSeek presents itself as an affordable alternative in the AI space, its high fail rate and susceptibility to misinformation highlight critical gaps that need addressing before it can be deemed a trustworthy source of information.
Source: https://www.newsguardrealitycheck.com/p/deepseek-debuts-with-83-percent-fail
