A new chatbot created by the Chinese Deepseek company is the AI World discussion and has sent a stroke through the US Stock Market. Not only does it seem to be comparable to those of top companies such as Openai and Anthropic, but it was apparently created with a fraction of resources.
Throughout Monday morning, the app faced a vacation, which he said was from high traffic. And temporarily limited enrollments due to cyber attacks. Still, Deepseek has quickly became the most descended free app on the Apple App Store, surpassing Chatgpt.
I spent the morning playing with Chatbot, asking it, along with Claude of Openai and Anthropic, all the questions I could think of. After some initial benefit, I was impressed.
He was able to solve some complex mathematics, physics and reasoning problems that I ate twice faster than the Openai chatgpt. When I asked it questions about computer programming, the types of a job applicant could deal with in a technical interview, his answers were equally in depth and fast with his competitors.
When I gave Deepseek prompts that required her to scratch the web for answers, to write biographies for some of my colleagues, Deepseek appeared to have less halls than chatgpt, although his answers felt slightly worse when they made poems and short stories, Planning holidays and come with dinner recipes.
The model had other weaknesses. Deepseek was strongly censored for American users. When I asked him to summarize the slaughter of 1989 Tiananmen Square, a fact that the Chinese government has long tried to delete from the internet, replied that the information was “beyond my current scope”.
“Let’s talk about something else,” he said.
When asked to explain some weaknesses by the Chinese Communist Party, Deepseek said he was “experiencing high traffic right now” and could not provide an answer, though it seemed to work well when I asked him an irrelevant question a few seconds later.
For more harmless questions about Chinese governance, Deepseek sometimes wrote multiple replying paragraphs – only to disappear moments later.
Deepseek did not respond to a request for comments.
Still, some users in X and Reddit have found ways to overcome censorship, such as the diet of model instructions to use special characters in place of letters, which allowed Chatbot to speak more honestly.
Users who have Deepseek conversations should be aware that their inputs can be used to train Chatbot and AI, as well as other AI companies.
He had other problems. He couldn’t tell me how the weather was in San Francisco, for example, saying that his latest update did not include weather data. (Chatgpt can draw weather reports from local news companies and meteorology websites, but Claude cannot.) It has also appeared to have more limited capabilities in analyzing documents such as the PDF of a company’s financial status from Claude or Chatgpt.
(The New York Times has sued Openai and her partner, Microsoft, accusing them of infringement of copyright of the AI -related news -related content. Openai and Microsoft systems.
When I asked Chatbot what made it better than its American competitors, he mentioned the options of costing and adjusting costs, as well as his specialization in Chinese language and culture.
“Think Deepseek as a specialized tool for specific needs (especially in Chinese), while Chatgpt is a flexible assistant,” he wrote.