OpenAI's new model Orion is scheduled to be here by December...a report that CEO Sam Altman is calling "fake news."
According to The Verge, OpenAI plans to release its next flagship model, codenamed Orion, by December.
However, the rollout will be a bit different than OpenAI’s two most recent releases, GPT-4o and o1. Instead of releasing Orion widely through ChatGPT, the AI juggernaut is allowing its close allies access first so that they can build their own products and features.
For example, OpenAI mega investor Microsoft will get Orion on Azure as early as November. This rollout seems to allude to a model more focused on enterprises than previous releases.
Now, for the model itself, The Verge is reporting that:
It is potentially up to 100 times more powerful than GPT-4.
It’s separate from the o1 reasoning model. Whereas that model was focused on inference-time compute (meaning it gets better the longer it thinks about problems), Orion seems to be the next stage in the GPT class of models, which relies on training-time compute. (They get better the more data it is trained on.)
OpenAI was using o1 to train Orion with artificial synthetic data.
OpenAI eventually wants to combine these models to create a model resembling artificial general intelligence (AGI).
Training for Orion finished in September, which OpenAI celebrated with a happy hour, and which CEO Sam Altman alluded to in this cryptic post that hints at the word “Orion”:
So, it sounds like we’re about to take the next step toward AI ascendance…except that Sam Altman is calling the report “fake news”, although he does clarify that “plenty of great stuff” is on the way.
As of now, it’s impossible to know who is telling the truth. Altman is obviously the CEO, but The Verge is one of the biggest news sites in tech, and reporter of record Kylie Robison is very plugged into the scene and has some big names vouching for her.
The only thing we can do is wait and see. But, if there’s one thing everyone involved can agree on, it’s that big things are coming soon.
This is fascinating news about OpenAI's upcoming Orion model! The focus on enterprise-level capabilities and the partnership with Microsoft for early access is a significant shift. If Orion truly is up to 100 times more powerful than GPT-4, it could revolutionize how businesses leverage AI. I'm particularly intrigued by the idea of combining training-time compute with reasoning models, as it could bring us closer to achieving AGI. I can't wait to see how this evolves!
It’s fascinating to see how OpenAI is moving towards more powerful, enterprise-focused solutions. I hope this new model lives up to the hype, as it sounds very promising!
Very Informative
Many people fear AI and new technology, believing it will take their jobs. However, the reality is somewhat different. The most developed countries on Earth have higher employment rates than less developed ones.
I’m eager to give it a try!
This comment was deleted 2 months ago