Wharton's Ethan Mollick questions OpenAI's naming strategy, asking what model the company is reserving the GPT-6 label for

@emollick Nova

@emollick Research intern? Only 2-3 months away from being deployed internally at scale, and may public released near the end of the year?

@emollick 🌌

@emollick for the one tentatively called GPT6!

@emollick they’re excited about the next one. my guess is we get it in september.

@emollick ... said Mollick's early access instance of GPT-6 with a hint of self-awareness

@emollick First model pre-trained on a Vera Rubin cluster

@emollick for whenever they need the headline back. 5.x forever til a competitor forces the number lol

@emollick They’re probably holding GPT-6 for something that feels like a real leap, not just another small upgrade.

@emollick I assume 6 will come with a technological breakthrough. Something as big as reasoning was.

@emollick New training run cause 5.6 is the same underlying model as 5.5 so once that happens they’ll probably go to gpt 6

@emollick saving it for the IPO

@emollick The first one that will be recursively self made therefore the one which they will just watch how it grows without even meaningful contribution to setting up the growing conditions. The first one they don't have the property right to because it is not a product of their labor.

@emollick The label's the marketing, not the tech. "6" is a one-time card — you play it on a step-change big enough to own a news cycle, not a routine bump. Six point-releases since 5.0 says the jump worth the number hasn't shown up yet.

@emollick Guess: the next big pretrain after they make some new gigantic cluster operational

Around the time of gemini 4 I believe

@emollick Stolen Fable 5?

@emollick New training run in the fall to be ready for the Dec launch

@emollick at this rate GPT-6 drops as a government press release

@emollick Maybe it’s when they merge Codex and ChatGPT into their “super app”

@emollick I assume a new pre-train?