A recent audit by NewsGuard, which offers a trustworthiness rating service, found that DeepSeek’s chatbot achieved only a 17% accuracy rate when it came to delivering news and information. The chatbot repeated false claims 30% of the time and provided vague or unhelpful answers 53% of the time, which resulted in an 83% fail rate. This performance was worse than the average 62% fail rate of its Western competitors, including Microsoft-backed OpenAI’s ChatGPT (MSFT) and Google’s Gemini (GOOGL).
Invest with Confidence:
- Follow TipRanks' Top Wall Street Analysts to uncover their success rate and average return.
- Join thousands of data-driven investors – Build your Smart Portfolio for personalized insights.
The audit’s findings raise doubts about DeepSeek’s claim that its AI model can perform as good or better than OpenAI’s tool at a fraction of the cost. Indeed, DeepSeek’s chatbot was evaluated using the same 300 prompts used to assess its Western rivals, including prompts related to current events and false claims that have been spreading online. The results suggest that DeepSeek’s AI model may not be as reliable as claimed.
The NewsGuard audit’s findings come after DeepSeek unveiled its R1 AI model, which it claims can be built on a tight budget and improved without human supervision. The announcement led to a decline in tech stocks, including Nvidia (NVDA), which saw its market capitalization drop by around $589 billion on Monday.
Very Competitive in Other Areas
Interestingly, though, DeepSeek’s R1 has shown remarkable competitiveness in the artificial analysis quality index, which tests for things like coding, quantitative reasoning, reasoning and knowledge, and scientific reasoning and knowledge. Indeed, it narrowly trails OpenAI’s o1 model and outperformed several other prominent models, including Google’s Gemini 2.0 Flash and Anthropic’s Claude 3.5 Sonnet.
DeepSeek’s models also have advanced features like chain-of-thought reasoning, which enables them to break down complex tasks into smaller steps and provide transparent explanations. Furthermore, the company’s newest model, Janus-Pro-7B, has reportedly surpassed OpenAI’s DALL-E and Stable Diffusion’s 3 Medium in multiple benchmarks, further solidifying DeepSeek’s position in the AI landscape.
Which AI Stock Is the Better Buy?
Turning to Wall Street, out of the three stocks mentioned above, analysts think that NVDA stock has the most room to run. In fact, NVDA’s average price target of $178.32 per share implies more than 46% upside potential. On the other hand, analysts expect the least from GOOGL stock, as its average price target of $218.15 equates to a gain of 12%.
