Technology giants like to boast of AI models of trillion of parameters that require huge and expensive GPU clusters. But Fastino is adopting a different approach.
The startup based in Palo Alto claims to have invented a new type of architecture model of artificial intentionally small and specific artificial intelligence. The models are so small that they are trained with low -end game GPU worth less than $ 100,000 in total, says Fastino.
The method is attracting attention. Fastino obtained $ 17.5 million in Seed funding led by Khosla Ventures, notoriously the first investor of Openi’s venture capital, Fastino only says to Techcrunch.
This brings the total financing of the startup to almost $ 25 million. He collected $ 7 million last November in a pre-sex round led by Microsoft VC Arm M12 and Insight Partners.
“Our models are faster, more accurate and cost a fraction to train, overlapping top models on specific activities,” says Ash Lewis, CEO and Fastino co-founder.
Fastino has built a suite of small models that sells to business customers. Each model focuses on a specific activity that a company might need, such as the drafting of sensitive data or the summary of company documents.
Fastino does not yet reveal early metrics or users, but says that his performances are amazed by the users of the former. For example, since they are so small, its models can offer an entire response in a single token, Lewis told Techcrunch, showing technology giving a detailed response immediately in milliseconds.
Techcrunch event
Berkeley, ca.
|
June 5th
Book now
It is still a little early to say whether Fastino’s approach will take hold. The space for the Enterprise is crowded, with companies such as Chere and Databrks that also propose to which excel in certain tasks. And the producers of SATA models focused on the company, including anthropic and teachers, also offer small models. Furthermore, it is not a secret that the future of generative artificial intelligence for Enterprise is probable in smaller and focused language models.
Time can say it, but a first vote of confidence from Khosla is certainly not hurt. For now, Fastino says that he focuses on the construction of a cutting -edge artificial intelligence team. The researchers of the best AI workshops is targeting who are not obsessed with the construction of the largest model or by beating the reference parameters.
“Our hiring strategy is very focused on researchers who perhaps have a contraniant thought process on how linguistic models are built right now,” says Lewis.
