On 22 October, TechCrunch reported that OpenAI made its Decisions API available to all customers, moving beyond the limited preview announced at its 29 September developer day. The API returns preset answers with confidence scores, as AI Affairs reported when it was introduced.
Nikunj Handa, product lead for OpenAI’s APIs, told TechCrunch that the company reconfigured ChatGPT Luna for faster, cheaper inference rather than training a dedicated decision model. He said response times fell from one to three seconds to 100, 120 or 180 milliseconds. OpenAI added voice and image support to the public version and placed inference resources geographically to reduce latency.
Andrew Hsu, chief technology officer at AI language tutor company Speak, said a trial cut the time between actions from 1.4 seconds to .13 seconds.
Moving through a language lesson could involve less waiting between actions if that reduction carried over to ordinary use.
TypeSafe’s Jev established the category of models that choose from supplied options and assign a confidence rating. TechCrunch reported that Jev ranks seventh and OpenAI’s model eighth on the AIM System One Index, and third and fourteenth respectively on the Artificial Analysis Decision Index.
