Tu Jinhao expressed many eye-catching viewpoints in an interview. He said that when AI answers questions, the “thought process” is more important than the answer. He would rather let AI slow down and search for more information than give a wrong answer casually. He also said that AI sometimes talks nonsense, but humans can also become confused, which is to some extent a manifestation of innovation.
If asked how he learned AI, he would be particularly honest and say, ‘Just use it more! Nowadays, AI is like a good friend, you can let it do whatever you want.’. Playing with it makes me feel interesting, and I want to figure out why it’s so smart. Then I search for information and read articles myself, and gradually learn. ”Nowadays, many media outlets are talking to him about AI “illusions” and how ordinary people play AI. He is definitely a representative figure of the post-00s generation playing AI.
Another author, Beverly, is not simple either. She studied computer science at Beijing Normal University for her undergraduate degree, and was recommended to study language and computer science at Peking University for her master’s degree. When she was in graduate school, she published 8 papers in a row at top NLP academic conferences around the world, of which 2 were first authors, directly becoming an academic star! Then she went to Ali Damo Academy and led her team to develop a multilingual AI model called VECO.
Later, she joined a quantitative investment company to research investment strategies using AI. In 2023, she joined DeepSeek and became the core developer of the DeepSeeker V2 model. After the release of DeepSeeker V2 in 2024, she said that this model has excellent Chinese proficiency, can compete with the world’s top closed source models, and is super cheap to use, with a price only one percent of GPT-4!
Her paper has been cited over 10000 times and has made significant contributions to the DeepSeek series of model research. A while ago, there was news that Xiaomi founder Lei Jun wanted to spend millions of dollars a year to poach her, and the title of “post-95s AI genius girl” suddenly spread.
In fact, behind DeepSeek, there are many young people like Tu Jinhao and Luo Fuli. For example, Yuan Jingyang, a master’s degree holder from Peking University, invented a super fast attention mechanism during his internship, which enables AI to process long articles faster; During his high school years, Gao Huazuo won first prize in a physics competition and later worked on research and development at DeepSeek; Daidamai joined the team after graduating with a PhD from Peking University and participated in the development of three generations of models throughout the process; Zeng Wangding studied AI at Beijing University of Posts and Telecommunications and also contributed a lot to the project. Liang Wenfeng once said,
The DeepSeek team consists mostly of graduates and doctoral students from prestigious universities, as well as some experienced young people. Together, they rely on creativity and professional knowledge to make DeepSeek stronger and stronger.
暂无评论内容