Technology giants are proud of the 1 trillion parameter AI model that requires a huge and expensive GPU cluster. But Fastino takes another approach.
Palo Alto -based startups say that they have invented a new type of AI model architecture by intentionally small and work. Fastino said the model is so small that he is being trained with a low -cost game GPU of less than $ 100,000.
This method is attracting attention. FASTINO has secured $ 17.5 million in seed funds led by Khosla Ventures, the first venture investor in Openai. Fastino speaks exclusively to Techcrunch.
As a result, the total funds of the startups are nearly $ 25 million. In November, he raised $ 7 million in the pre -seed round led by Microsoft’s VC ARM M12 and Insight Partners.
Ash Lewis, the CEO and co -founder of Fastino, said, “Our model is faster and accurate and it has been categorized to classify it while surpassing the flagship model for a particular task.
FASTINO has built a small model product line for enterprise customers. Each model focuses on specific tasks that the company may need, such as modifying sensitive data or summarizing company documents.
Fastino says that it has not yet released initial metrics or users, but performance will surprise the initial user. For example, because it is too small, the model can deliver the entire response with a single token, and Lewis speaks to TechCrunch, and the technology provides a detailed answer at once.
TECHCRUNCH event
Berkeley, California
|
June 5th
Please make a reservation now
It is still a little early to tell if Fastino’s approach will be taken. Companies such as Cohere and Databricks are crowded with enterprise AI space that promotes excellent AI for specific tasks. In addition, enterprise -oriented SATA model manufacturers, including humanity and miss, also offer small models. In addition, it is not a secret that the future of the creation AI for companies is in a smaller and more concentrated language model.
Time can be said, but Khosla’s initial trust votes are certainly not painful. Fastino is currently focusing on building a state -of -the -art AI team. It is aimed at TOP AI LABS researchers who are not obsessed with building the largest model or hitting the benchmark.
Lewis said, “Our hiring strategy focuses on researchers who have the opposite thinking process of how language models are being built.
