Will Brown called Persimmon one of the most exciting recent model releases from a new Western lab and praised the utility of realistic user simulators. He also speculated that the technology might be generating revenue through a private API used in reinforcement-learning environments.
Niloofar Mireshghallah pushed back on that commercial interpretation. She said the user simulators are not being commoditized, research access is currently very limited and there is no intent to monetize them because of ethical and misuse concerns.
Why realistic user simulators matter
Brown suggested this approach could help frontier models deal more effectively with everyday interactions, including corrections and half-formed thoughts.
Alexis Ross argued that user models and simulators like Persimmon could enable new training methods and new kinds of AI models. That is a view about the technology’s potential, not evidence that any particular training program or product has been announced.
Access remains deliberately narrow
The discussion does not establish that Persimmon is offering a commercial API. Mireshghallah’s statement points in the opposite direction: limited research access and no current plan to monetize.
Eric Zelikman said there may be a way forward, but that it should be approached with substantial thought and care. For now, the clearest stated boundary is that access remains limited while ethical and misuse risks are considered.
Comments