Modifications within the agent tech stack vary from one thing so simple as updating the underlying AI mannequin’s model, to transferring from a closed-source to an open-source mannequin or altering the database the place agent information is saved, he notes. In lots of instances, changing one part within the stack units off a cascade of modifications downstream, he provides.
“While you go to an open-source mannequin that you simply run by yourself server, your entire infrastructure modifications, and it’s important to take care of quite a lot of belongings you weren’t coping with earlier than, and then you definitely would possibly go, ‘That was truly worse than we anticipated,’” Northcutt says. “So that you return to a special mannequin, however then you definitely swap to cloud, and the cloud API is definitely completely completely different than the OpenAI API, as a result of they don’t seem to be in settlement.”
Cozmo AI, a voice-based AI supplier, has additionally noticed a sample of frequent modifications in agent tech stacks, says Nuha Hashem, cofounder and CTO there. The Cleanlab survey matches the churn Cozmo sees throughout regulated environments, she says.
Leave a Reply