Is the model Orion the next big thing? Maybe, maybe not. Nobody seems to know more than the name right now.
“We don’t have plans to release a model code-named Orion this year,” a spokesperson told TechCrunch via email. “We do plan to release a lot of other great technology.”
Okay then. A bunch of new stuff is coming our way, hooray for that. People do like shiny and new. But what about Orion?
Orion, a step up from OpenAI’s current flagship, GPT-4o, is reportedly trained in part on synthetic training data from o1, the company’s “reasoning” model. OpenAI plans for the foreseeable future to continue developing new “GPT” models alongside reasoning models like o1, which it sees as addressing fundamentally different use cases.
https://techcrunch.com/2024/10/25/openai-says-it-wont-release-a-model-called-orion-this-year/
I’m not really sure what we’re supposed to think about the model being trained on synthetic data. What will the end user experience be with synthetic data inside the model? I mean hopefully better right. But that might not always be the case as people have shown o1 has some bizarre suggestions for people. Are those suggestions like using something like 4 ovens to cook a meal going to get filtered out or are they going to become part of the reasoning of the new model?
o1 is not perfect and it makes sense that a new model would be in the works. I guess the only question is what is it called, right now everyone is guessing Orion. But regardless of the name, there will be a model.
I’d like to see a new DALL-E model however. It’d be nice to get a model that can consistently spell and count. They did seem to improve the issue with too many or too few fingers but there’s plenty of other counting issues left to solve… like when you ask for A picture of a car and you get THREE cars. Maybe I’ll get lucky and that’ll be some of the new technology they mentioned.
Anyway Orion is not on the calendar for this year… that’s what we should all take away from this. Maybe next year folks!