Trajectory gets $45M to let experts retrain AI models
Trajectory says that lawyers, doctors and other experts can retrain open-source models near automatically, but it gave no test results.
Trajectory, a startup of 18 persons, got $45 million in a Series A funding at a $300 million valuation. Trajectory gives a platform that retrains an AI model from feedback on its work, to do one task more correctly. The platform is open to all persons at this time, and Sequoia is the primary investor.
How it works
A business starts with an open-source model and connects the platform to its software. The platform collects feedback on how the model does and retrains it. Co-founder Ronak Malde says that experts can do this near automatically, with no research team and no consultants.
Who uses it
Harvey and Rogo use the platform to train models for law and finance. Other customers are:
- Decagon
- Mercor
- Nvidia
What is not known
The company gave no test results. Co-founder Michael Elabd says that training with persons in control is more safe than AI that works without persons.
Insight: Two ways to put expert skill into AI
iMerit helps AI developers get and use more than 30,000 experts to test and tune models. Trajectory gives the experts of a firm software to do it. This reads as a difference between experts for hire and a loop that the firm controls. The one that controls the loop can control the results. Find out which one your vendor gives.
Sources
Posted