In recent days, the Chinese artificial intelligence company DeepSeek has stirred up a storm due to its claim of being low-cost and its performance comparable to the leading AI models from the United States’ OpenAI. However, some financial experts in Taiwan have raised doubts about whether there’s something fishy going on. In China, some professionals have also questioned DeepSeek’s capabilities, speculating that the company may be relying on plagiarism to make significant strides.
DeepSeek, a previously obscure technology company in China, has gained attention for its artificial intelligence model, DeepSeek, which is touted as a large language model developed with low-cost hardware that rivals the most advanced AI models from the U.S. The company’s products have received both praise and skepticism, with some questioning the integrity of DeepSeek’s offerings.
According to a Chinese technology blogger, DeepSeek has shocked the U.S. tech community with its claims of advancements that could potentially threaten American AI. However, suspicions have been raised about the authenticity of DeepSeek’s achievements, with some experts suggesting that DeepSeek may have used distillation techniques on the GPT-4 model. This shortcut method may lead to advancements but could hinder genuine progress and innovation in the long run.
The blogger further explains that comparing DeepSeek and GPT-4 reveals striking similarities in their capabilities, indicating a high likelihood of content distillation. While this learning method may yield quick results, it could hinder students (or in this case, AI models) from surpassing their teachers (or original creators) once they are no longer provided with the answers.
Another blogger, who is an AI entrepreneur in China, tested DeepSeek against ChatGPT in solving logical reasoning questions and found discrepancies. This raises concerns about DeepSeek’s accuracy and reliability compared to other AI models. The entrepreneur emphasizes that despite narrowing the gap with OpenAI, surpassing it seems unachievable, as even DeepSeek’s founder refrains from claiming superiority over OpenAI.
Ruby Scanlon, an assistant researcher at the Center for a New American Security (CNAS), highlighted DeepSeek’s ability to create products with lower-end hardware and cost-efficient methods, likely through model distillation techniques. By focusing resources on critical components of a model, companies like DeepSeek can produce better solutions while minimizing costs.
DeepSeek, a subsidiary of the Chinese technology company MagicQuant, has introduced a range of AI language models that have garnered attention globally. The company’s cost-effective approach and impressive performance have caused a stir in the tech industry, leading to significant drops in the stock prices of American tech giants.
Despite DeepSeek’s claims of cost-effectiveness and performance comparable to OpenAI, external testing has revealed issues with the model’s political biases, particularly in avoiding sensitive topics related to the Chinese government. Users have reported instances where DeepSeek either deflects questions or provides politically inclined answers, raising concerns about censorship within the AI.
Critics have highlighted instances where DeepSeek’s responses differ significantly from those of other AI models, leading to accusations of political manipulation and bias within the technology. This revelation has sparked debates about the trustworthiness and objectivity of AI algorithms in addressing sensitive topics and issues globally.
The diverging opinions and controversies surrounding DeepSeek’s development and functionality underscore the complexities and challenges in creating ethical and unbiased AI systems. As the AI landscape continues to evolve, scrutiny over the authenticity and integrity of AI technologies like DeepSeek will likely intensify, prompting further investigations and discussions within the tech community.