Interesting statements by one of Nvidia’s managers about artificial intelligence

Jonathan Cohen, Vice President of Applied Research by Nvidia, believes that in the field of artificial intelligence research, the main restrictive factor, the amount of computational resources is available.

“In the present age, the most valuable asset for any artificial intelligence researcher is his access to graphic processing units (GPUs), and this is as important as any other organization in Nvidia,” said Cohen in an interview with Nvidia Developer.

Cohen was in charge of a group responsible for the development of the family of Nvidia’s large language models (LLM). These models, introduced in March of this year, indicate the company’s entry into the field of intelligence artificial intelligence systems.

Cohen said that the speed of forming these models was remarkable and surprising, and the process did not last more than one to two months. “Many scholars have agreed to provide our own computational resources so that we can teach and move Llama and Nemotron models as fast as you saw,” he said.

Cohen also attributed the high speed of the development of these models to the dominant and dominant culture of Nvidia, which prioritizes large and large projects, regardless of the current and current goals of different teams.

“How do you form a team to do something that has never been experienced before?” Part of our organizational culture has been something that we call “congestion” and is recognized in a matter of importance. In this situation, all managers are thinking about whether this new job is more important than the thing that all the teams are doing now? “

If a manager recognizes that this new project has a higher priority, it can discharge employees from their previous tasks and assign them to a new priority. “Finally, the Llama Nemotron project was a very widespread inter -team effort,” Cohen said. “We had people from all over the organization who worked together without any formal and defined organizational structure.”

According to Cohen, the development of Llama Nemotron required a set of sacrifices, both in terms of the necessary computational capabilities and in terms of specialist human resources, and the company’s staff was able to abandon their personal interests and individual goals in line with the overall interest and purpose of the organization.

“Seeing this amount of cooperation and sacrifice was really great and admirable,” he said. There was an extraordinary leadership. “Many sacrifices were made by people, and many sacrificial decisions were made that eventually led to the formation of the project, which is really amazing.”