In a shocking turn of events, DeepSeek’s R1 chatbot has surpassed OpenAI’s ChatGPT as the top-ranked app on Apple’s US App Store. But what sets these generative AI tools apart?
The recent release of DeepSeek‘s R1 chatbot has sent shockwaves through the tech industry, temporarily dethroning ‘OpenAI‘s ChatGPT‘ from the top spot on Apple’s US App Store. This Chinese startup’s innovative approach to training efficiency has raised questions about the need for high-powered chips and expensive data centers.
A Breakthrough in Model Training Efficiency
DeepSeek‘s R1 model is a powerful generative AI tool that can be accessed through the company’s smartphone apps, desktop website, or third-party services like ‘Perplexity Pro.’ The app is completely free to use, with no subscription fees or paywalls. This is in contrast to ‘OpenAI’s ChatGPT,’ which requires a $20-a-month subscription fee for its basic version and $200-a-month for the most capable model.
Comparing DeepSeek to ChatGPT
While DeepSeek‘s R1 model is comparable to ‘OpenAI’s o1 ‘reasoning” model, it still suffers from common weaknesses found in generative AI tools. Hallucinations, where a chatbot makes up information in its responses, are rampant in both DeepSeek and ChatGPT. Invasive moderation and questionably scraped material also plague these platforms.
Testing the Limits of DeepSeek
I tested the DeepSeek app for several hours, generating various outputs to assess its capabilities. While I was impressed by the model’s ability to gather links from the web and provide helpful answers, I was disappointed by its tendency to make up information. For example, when asked about my background without searching the web, the bot confidently stated that I’m a longtime tech reporter at ‘The Verge’ – not true!
A Research Breakthrough with Room for Improvement
DeepSeek‘s breakthrough in model training efficiency is significant, but it doesn’t address the fundamental issue of hallucinations. Many of the outputs I generated included blatant falsehoods, which raises concerns about the accuracy and reliability of these AI tools.
Multimodal Capabilities and Transparency
While DeepSeek has earned praise for its locally available open weights, allowing users to adjust the model’s capabilities, it still lacks transparency about what data was used for training. The startup’s lack of a communications team or press representative makes it difficult to obtain information about their methods.
Transparency refers to the quality of being clear and honest in communication, actions, and decisions.
It involves providing accurate and timely information to stakeholders, including employees, customers, investors, and the public.
Transparency is essential for building trust, accountability, and credibility in business, government, and other organizations.
Research shows that transparent companies outperform non-transparent ones by 10% (PwC).
In fact, 85% of consumers trust brands that prioritize transparency (Edelman).
Conclusion
The success of DeepSeek‘s R1 release is unlikely to mark the end of American AI leadership. Instead, it may transform how we think about AI tools and inspire companies like ‘OpenAI‘ to borrow ideas from DeepSeek‘s architecture. As with ChatGPT‘s original release, this high-powered free model will likely redefine the shape of the current AI industry.
