Open models for closed: AI leaders of GM, Zoom and IBM are webs to use the institution
Want more intelligent visions of your inbox? Subscribe to our weekly newsletters to get what is concerned only for institutions AI, data and security leaders. Subscribe now
Decide on Artificial intelligence models It is a major technical decision and is a strategic decision. But choosing open, closed or mixed models all have bodies.
During his speech at the year VB Transform for this year, architectural engineering experts from General Motors, Zoom and IBM discussed how their companies and customers think about choosing the artificial intelligence model.
“There is a lot of noise with every issuance of new models and every time the leaders change,” said Barack Tofsky, who in March. The first Amnesty International officer in General Motors said. A long time before the leaders of the prevailing discussion, TUROVSKY helped the first launch Language model (LLM) and calling the roads that led the model weights, open source training data and training data to large penetrations.
“Perhaps this is frankly one of the biggest breaches that helped Openai and others to start at all,” Tofsky said. “So it is actually a funny story: I actually helped the source of creating something closed and now it may return to openness.”
Decision factors differ and include cost, performance, confidence and safety. Turovsky said that institutions sometimes prefer a mixed strategy – using an open model for internal use and a closed model of production and confronting customers or vice versa.
IBM AI Strategy
Armand Ruiz, Vice President of IBM from AI, said that IBM initially started her platform with her Special llmsBut then I realized that this would not be enough – especially as the most powerful models reach the market. Then the company expanded to provide integration with platforms such as Huging Face so that customers can choose any open source model. (The company recently appeared as a new model gate Which gives institutions an interface programming interface to switch between LLMS.)
You choose more institutions to buy more models from multiple sellers. When Andresson Horwatz to survey 100 information managers, 37 % of the respondents said they use 5 or more models. Last year, only 29 % used the same amount.
Ruiz said the choice is the key, but sometimes it creates a lot of confusion. To help customers in their approach, do not worry IBM a lot about the LLM they use during the concept or experimental stage; The main goal is the feasibility. Only later they start looking at whether a model will be distilled or customized based on the customer’s needs.
“First, we try to simplify all this paralysis analysis with all these options and focus on the state of use,” said Ruiz. “Then we discover what is the best way to produce.”
How to enlarge the enlargement of artificial intelligence
Zoom Cto Xueong Huang said among Zoom customers between two compositions of her companion of artificial intelligence. One of them includes the company’s LLM followers with other larger basis models. Another confrontation allows customers interested to use many models to use the Zoom model only. (The company also made a partnership with Google Cloud to adopt an agent of an agent for an artificial intelligence agent for the functioning of institutions.)
Huang said that the company made its small language model (SLM) without using customer data. On two billion parameters, LLM is actually small, but it can still outperform the other industry models. SLM works better in complex tasks when working alongside a larger model.
“This is really the power of a hybrid approach,” said Huang. “Our philosophy is very clear. Our company leads the road like Mickey Mouse and dancing in the elephant together. The small model will perform a very specific task. We do not say that a small model will be good enough … Mickey Mouse and the elephant will work together as a team.”
[publish_date
