OpenAI could launch its next generation model in December


OpenAI plans to launch its next-generation frontier model, codenamed Orion and rumored to actually be GPT-5, in December, according to an exclusive report from The Verge. However, OpenAI boss Sam Altman is already responding.

According to “sources llano with the plan,” Orion will not initially be released to the universal public, as previous GPT-4 variants were. Instead, the company intends to deliver the new model to selected companies and partners, who will then use it as a platform to develop their own products and services. This is the same strategy that Nvidia is following with its NVLM 1.0 family of large language models (LLM).

What's more, The Verge reports that Microsoft plans to host the new model starting in November. There is no confirmation yet that Orion will be called GPT-5 when it launches, although its engineers reportedly consider the model to be the successor to GPT-4.

Within hours of the report's release on Friday, Altman coyly denied the claims and stated that “fake news is out of control.” However, he did not outright deny that Orion would be released in December, nor did he cite any specific aspect of The Verge's story as factually inaccurate.

He hasn't explained his position since posting that tweet, leaving us confused both about the meaning of his statement and his company's eventual plans for Orion.

This report comes just over a month after OpenAI's surprise launch of Project Strawberry, officially known as 01 preview and 01-mini. Its “reasoning” architecture is designed to deduce answers like a human would and accurately solve complex questions on various topics, including science, coding, and mathematics, faster than a person.

However, the 01 models have not been received with the same degree of fervor that GPT-4 received, in part because the new models have limited functionality, are unable to upload files or analyze images, and as VentureBeat notes, are expensive to install. operate for OpenAI.

In the lead-up to the release of the 01 preview, Altman posted a series of cryptic tweets featuring the fruit. He seems to be doing the same with Orion. In September, just as the company was reportedly finishing Orion training using synthetic data generated by 01, Altman sent out a striking tweet about visiting the Midwest. As The Verge points out, the dominant constellation in the winter sky of the Northern Hemisphere is, as you may have guessed, Orion.

Rumors about GPT-5 have been flooding the Web for many months, basically since the day GPT-4 was released in March 2023. Reports initially pegged the release date sometime in summer 2024, but merienda that passed, the goal has been moved to this fall. On the other hand, former CTO Mira Murati said in an interview last June (before leaving the company) that the “next generation” model would not come out for a year and a half.

So while a late December launch, coinciding with ChatGPT's second anniversary, seems likely, it's equally likely that it won't arrive until 2025 given how inaccurate all the predictions have been so far.





Qmx">Source link

Leave a Comment