Given a lot of political pressures, i think it's somewhat plausible for there not to be any model called GPT-5, that openAI just releases upgraded ChatGPTs. Maybe they work only on multimodal models from now on and never release any pure large language model again, etc. and discontinue the GPT series.
Will there be a large language model titled officially as GPT-5 released by OpenAI? It can be a multimodal model called GPT-5, or it can be a language only model called GPT-5. The question is whether they will discontinue this particular series and do something different, as ex: Gobi, arrakis etc suggest.
In case OpenAI splits up etc, i will make changes at the time to reflect the resolution criteria better.
i think it's less likely that they release it named as a GPT-5 if they're moving away from pure language models, but if they do release a multimodel GPT-5, then it counts
Related questions
@mattyb what if they just go for gpt X or gpt, and then stop updating the number but change it in backend.
@chrisjbillington what's your thought process here? You seem very confident that the naming convention will not change
@Seeker I'm aware Sam says "we'll release a great model this year, not sure what we'll call it" or similar, but haven't listened to that part myself yet. Nonetheless I think any indecision is more likely to be about whether to call it GPT-4.5 or GPT-5, or something like that, not whether to break the naming convention.
Ultimately they have massive brand recognition with the "GPT-N" naming scheme, and would be absolute fools to do away with it. I don't think they're that foolish.
@Seeker anything else relevant in the podcast? Entirely possible there's more evidence there than I'm aware of - I'm currently just betting on the basis that the "we don't know what we'll call it" line is all there is (and that people are interpreting it incorrectly).
Am listening to it myself and will find out soon enough, though.
@chrisjbillington I mean... I think they have massive brand recognition with ChatGPT, more so than the numbering system with GPT-x. But I also think that those names are objectively bad and I think they'd be "fools" to keep iterating GPT indefinitely. I just don't think OpenAI's brand appeal is really tied up that much in the GPT-x name. I think everyone knows OpenAI's models are the best, and the name of that is not going to influence it much. It's not like people will cancel their subscriptions when the model name changes from GPT-4-turbo to "Bella" or whatever.
Also, the podcast wasn't what swayed me, I just hadn't seen this market before it. I would've bet this down months ago. The podcast quotes didn't really shift my thinking substantially.
Really what is affecting my thinking is that they named Sora an anthropomorphic name, unlike DALL-E 2/3 or GPT-3.5/4 or ChatGPT, which makes me think the company is moving more towards well-named models rather than acronyms.
But I'm just surprised you're confident enough to bet this up to nearly 90%!
@benshindel yes! The naming shift is somewhat noticeable, hopefully I can resolve the following market soon
@benshindel DALL-E is kind of doubly anthropomorphic, it's a pun on WALL-E and Salvador Dalí, so I don't think there has been a shift.
It's true that ChatGPT has more recognition than any individual model, but they still need to market the models to get people to pay for ChatGPT Plus, and right now "GPT-5" connotes "as big a jump as from GPT-3 to GPT-4" which is an amazing connotation to have in a name right out of the gate.
We're up to iPhone 16 or whatever, they're not changing their naming. Incrementing a number on a well-recongised product has connotations of being a serious company knowing what you're doing and producing consistently reliable products. Changing it risks looking weak and desperate.
The fact that "GPT" is a bad name isn't really relevant I don't think. Once names have recognition, what they would sound like to new ears no longer matters. Names come to be associated with the thing they represent, and other connotations fade away.
Anyway, we'll see!
@firstuserhere I strongly suspect they're just waiting on availability of H100s. Starting a run just as the hardware generation is changing would be weird.