https://x.com/tsarnick/status/1803901130130497952
"Mira Murati: GPT-3 was toddler-level, GPT-4 was a smart high schooler and the next gen, to be released in a year and a half, will be PhD-level"
transcript is less direct:
And then in the next couple of years, we're looking at PhD-level intelligence [...]
[Interviewer] Meaning like a year from now?
Yeah, a year and a half, let's say.
Ways in which I could be wrong:
The new foundation model they're training is actually already GPT-6; they are going to go easy on the safety training; they started training a while ago and only announced it a month ago; they will bring out an improved GPT-4 and call it GPT-5 (sub-case of this: they figure out some STaR variant that makes the model good enough to be called GPT-5, they feel strong pressure by Anthropic et al. & bring out an underwhelming model).
.
@TobiasH As of 1708866021 Gigacasting is liking Tobias' comment, so I view this as a declaration that a successor to GPT-4 that is not called GPT-5 would not be sufficient to resolve this market YES.