AI training startup Trajectory raises $45 million in a round led by Menlo Park's Sequoia

Trajectory, a startup that helps businesses keep improving their AI models after they are built, has raised $45 million, TechCrunch reported Monday.  The round, which values the 18-person company at $300 million, was led by Menlo Park venture firm Sequoia Capital, with Nvidia, Bessemer Venture Partners and Bain Capital Ventures also investing.

The company also opened its software to the public Monday, according to its website.

The problem it tackles

AI models such as ChatGPT learn during a long, expensive training process and then mostly stop learning.  If a model keeps making the same mistake on a law firm's contracts, it does not fix itself.  Companies usually hire AI researchers or consultants to retrain, or "fine-tune," the model with extra examples.

Trajectory sells a tool for what researchers call continual learning.  A business connects an open-source AI model, one whose inner workings are freely available, to its own app.  Trajectory records signals such as when workers edit, correct or retry the AI's answers, then uses that feedback to retrain the model.  Each update must pass the customer's own tests and approval before it goes live, the company says.

what the economy values is someone who worked on the job for 30 years

Who is behind it

Trajectory was founded about a year ago by three Stanford University classmates:  Malde and Michael Elabd, who worked at Google DeepMind, and Arjun Karanam, who worked on AI at Apple.  Customers include legal AI company Harvey, finance AI company Rogo, customer-service AI company Decagon, Mercor and Nvidia.

Rivals include Thinking Machines Lab, the San Francisco startup led by former OpenAI technology chief Mira Murati, and Applied Compute, which TechCrunch said was valued at $3.25 billion last month.

Source:  TechCrunch; Trajectory.