One observation I found interesting was Altman said that the OA API+Codex is profitable, but not profitable enough for the ‘next generation’ of models.
It’s what I would have guessed at the estimated revenue numbers, but it’s good to know. It also means that they’re probably going to seek more VC (from MS?) for continuing upgrading & scaling, since we’re seeing increasing competition. (Google alone has posted at least 2 papers so far using LamDA which is at a comparable scale, for coding as well, and their mysterious ‘Pathways’ model is apparently multimodal and even larger.)
One observation I found interesting was Altman said that the OA API+Codex is profitable, but not profitable enough for the ‘next generation’ of models.
Yeah, but he said “at current revenue”, which is kind of unsurprising?
Codex/Copilot is all of 6 weeks old and GPT-3 based apps that see real usage yet are where?
It’s what I would have guessed at the estimated revenue numbers, but it’s good to know. It also means that they’re probably going to seek more VC (from MS?) for continuing upgrading & scaling, since we’re seeing increasing competition. (Google alone has posted at least 2 papers so far using LamDA which is at a comparable scale, for coding as well, and their mysterious ‘Pathways’ model is apparently multimodal and even larger.)