OpenAI introduced its subsequent main AI mannequin Saturday, and it did so in the third paragraph of a blog post called “Ten advances in arithmetic and theoretical laptop science.”
The mathematics outcomes the publish is touting, OpenAI writes, “had been achieved by an inside model of Astra, our subsequent main mannequin.” So there you go. It seems like after GPT-5.6 Sol comes both GPT-5.6 Astra, or GPT-6 Astra, or—who is aware of?—simply “Astra” and the entire GPT half will get scrapped? It’s not spelled out.
OpenAI’s naming conventions recommend that this might be one other GPT-5.6 launch. There’s a GPT-5.6 Terra, which, as corny Latin-knowers are properly conscious, means “earth,” Luna which is Latin for “moon,” and Sol, which is Latin for “solar.” Astra means “the stars.”
In accordance to an anonymously sourced story in the Information, Astra boasts the capacity to do “long-running” work. CEO Sam Altman was, the report claims, in Washington, D.C. over the previous week, demoing the mannequin to federal officers.
Gizmodo requested OpenAI on Saturday for the official identify of the mannequin. We additionally requested OpenAI about the relationship between this mannequin and one other OpenAI mannequin solely vaguely described in one in every of that firm’s weblog posts. We did not obtain a reply.
The “unprecedented cyber incident” coated in that July 21 blog post is already notorious. An entity described as a “mixture of OpenAI fashions — together with GPT‑5.6 Sol and an much more succesful pre-release mannequin, all with lowered cyber refusals for analysis functions” compromised the AI useful resource depository Hugging Face throughout a mannequin analysis train that was supposed to stay inside OpenAI.
A later replace to the weblog publish clarified that the unreleased and unnamed mannequin concerned in the incident was an “internal-only analysis prototype and was by no means meant for public launch,” and that it had been “deactivated, encrypted, and restricted.” So to be clear, Astra is not the mannequin that broke into Hugging Face.
Gizmodo requested OpenAI on Saturday to extra totally make clear the relationship or lack thereof between Astra and the never-to-be-released mannequin. We did not obtain responses in time for publication, however will replace if we obtain a clarifying reply.
As for the weblog publish about math, it comes with a paper. There are ten proofs coated, protecting such matters as the “asymptotic power of the Cohn–Elkies linear program,” for sphere-packing, which OpenAI purports to be “decided precisely.” That sounds very cool, however I’m simply the man who blogs nights and weekends for Gizmodo.
For what it’s price, OpenAI published a mathematical disproof back in May, purportedly solved by an unnamed OpenAI mannequin. For the most half, math people wrote about the mannequin’s work approvingly, however didn’t appear fully knocked out.
For example, Harvard mathematician Melanie Matchett Wooden wrote that OpenAI’s proof was, “a phenomenal utility of quantity principle to a pure, concrete query,” however she additionally stated the downside, described by OpenAI as “a central conjecture in discrete geometry” was nothing she had ever heard of before. She added:
“This end result does not present us all the occasions AI has claimed to have a proof of one thing and been fallacious. With out that context (which many people have simply from private expertise), it is additionally straightforward to draw incorrect conclusions about the present state of AI and analysis arithmetic.”
Disclaimer: This article is sourced from external platforms. OverBeta has not independently verified the information. Readers are advised to verify details before relying on them.
