- The NVIDIA CEO, the CEO of NVIDIA, said investors have misrepresented Deepseek.
- The large language models were built in Deepseek with weaker chips, and comic markets in January.
- Huang stressed the importance of training artificial intelligence in an previously recorded interview issued on Thursday.
Jensen Huang, CEO of NVIDIA, said at a virtual event that was broadcast on Thursday, Jensen Huang, CEO of NVIDIA, said at a virtual event that was broadcast on Thursday, CEO of Nvidia Jensen Huang said in a virtual event that was broadcast on Thursday that investors had removed The wrong message of Deepseek’s progress in AI.
Deepseek, a Chinese company of artificial intelligence owned by the hedge fund, has released an open -source competitive model called R1 in January. The company said that the Grand Language model on the R1 was built with weaker chips and a small part of financing prevailing artificial intelligence models.
Investors’ reaction to this news was by selling NVIDIA shares, which led to a loss of $ 600 billion in the market value. Huang himself temporarily has approximately 20 % of his net wealth in guidance. The stock has since regained a lot of its lost value.
Huang said in an interview registered on Thursday, which was produced by NVIDIA partner NVIDIA partner NVIDIA partner NVIDIA partner NVIDIA partner NVIDIA NVIDIA partner NVIDIA partner NVIDIA partner NVIDIA partner NVIDIA NVIDIA partner NVIDIA partner NVIDIA NVIDIA partner NVIDIA partner NVIDIA partner partner NVIDIA NVIDIA partner and part of the new software platform in DDN.
Investors raised questions about whether trillion needed to spend on Amnesty International’s infrastructure by major technology companies, if a lower computing force is needed to train models. Jensen said that the industry still needs a computing power of post -training methods, which allows the artificial intelligence models to extract conclusions or make predictions after training.
As post -training methods grow and diversify, the need for computing energy provided by NVIDIA chips will also grow.
“From the point of view of the investor, there was a mental model that the world was a pre -training and then reasoning. The reasoning was: Amnesty International asked a question, and I immediately got an answer,” he said in the event on Thursday. “I don’t know who is his mistake, but it is clear that this model is wrong,”
Preparation is still important, although post -training is “the most important part of intelligence. This is where you learn to solve problems.”
He said that Depsik’s innovations stimulate the world of artificial intelligence.
“It is very exciting,” said Huang.
NVIDIA spokesman addressed the market’s reaction with written data with a similar effect, although Huang has not yet indicated this topic until Thursday happened.
Huang was defending the growing anxiety that scaling the model in trouble for several months. Even before Dibsic pushed public awareness in January, reports that the typical improvements in Openai were slowing down the devastating doubts that the mutation of artificial intelligence may not fulfill its promise – so NVIDIA will not continue to take advantage of the same rate.
In November, Huang confirmed that scaling is alive and in good health; It simply turned from training to reasoning. Huang also said on Thursday that post -training methods are “really intense”, and models will continue to improve with new thinking methods.
Huang Deepseek’s comments may be a preview of the first NVIDIA profit call for 2025, scheduled for February 26. Deepseek has become a common theme for discussion on companies ’profit calls via a technology spectrum, from Airbnb to Palantir.
The NVIDIA AMD competitor was asked the question earlier this month, and the CEO Lisa Soo “Good to adopt artificial intelligence,” said Dibsic.
