The DeepSeek logo on a mobile device. (Image credit: Bloomberg via Getty Images)
Chinese artificial intelligence (AI) technology firm DeepSeek has unveiled a new image generator shortly after its successful chatbot generated significant interest in the tech industry and stock markets.
DeepSeek has gained attention for its partially open-source AI models, which compete with OpenAI's ChatGPT despite having significantly lower development costs.
DeepSeek announced on Monday (January 27) that the latest version of its free Janus image generator, Janus-Pro-7B, performed better than OpenAI DALL-E 3 and Stability AI's Stable Diffusion in benchmark tests, Reuters reported. However, Artificial Analysis, which compares the performance of different AI models, has yet to independently evaluate DeepSeek's Janus-Pro-7B against other models.
The announcement of the image generator comes at a significant moment for DeepSeek and the entire field of AI technology. First of all, DeepSeek’s free chatbot became the most downloaded app in Apple’s US App Store, beating out ChatGPT. That same day, the company announced that it was temporarily restricting user registration due to a major cyberattack, although existing users continue to have access to the system, CNBC reports.
The stock market also reacted to the success of the low-cost chatbot DeepSeek on Monday. Leading AI chipmaker Nvidia lost $589 billion on the stock market, the largest one-day loss in U.S. history (though the company is still valued in the trillions of dollars).
DeepSeek’s AI models have made waves in the tech industry because they require fewer computing resources to run than traditional algorithms, making them more cost-effective. The U.S. has restricted imports of high-quality AI computing chips to China, prompting the DeepSeek team to develop smarter, more energy-efficient algorithms that are less resource-hungry, Live Science previously reported.
Janus-Pro-7B is a free model that can analyze and generate new images. Like other image generators, users enter a text description of the image they want, and the generator creates it. Improvements to the model were achieved through new training methods, better data quality, and a larger model size, according to a technical report seen by Reuters.
DeepSeek fed the model 72 million high-quality synthetic images and compared them with real data, which reportedly allows the Janus-Pro-7B to produce more aesthetically pleasing and stable images than competing generators.
DeepSeek uses “open weight” models, meaning users can view and modify the algorithms even though they don't have access to the training data.
Patrick PesterNavigate Social LinksPopular News Author
Patrick Pester is a well-known news writer for Live Science. His work has appeared on other science platforms such as BBC Science Focus and Scientific American. Patrick moved into journalism after working in zoos and wildlife conservation earlier in his career. He received a Master's Excellence Scholarship to study at Cardiff University, where he completed an MA in International Journalism. He also has a second MA in Biodiversity, Evolution and Conservation in Action from Middlesex University in London. When he's not writing news, Patrick is researching the sale of human remains.
You must verify your public display name before commenting.
Please,
Sourse: www.livescience.com
