Gary Marcus

Rank 22 of 47
|
Score 35
8/2/2024, 4:47:30 PM
X
In reply to:
Nathan Cahill
@nathanmcahill
·
409d
@GaryMarcus FWIW I agree with you re: saturating frontier model performance.

TBH I think GPT4 was a miracle ahead of its time as well. The team at OpenAI is (was?) so legendary for their intuition on what to include in their huge Yolo training runs.
Gary Marcus
@GaryMarcus
·
411d
The thing that really gets me is that tech people are supposed to understand … math
Gary Marcus
@GaryMarcus
·
411d
👇 Sure, tell me we aren’t reach a period of diminishing returns.

Giant leap from Chinchilla to GPT-4; no comparable leap since, even though there are now 𝘷𝘢𝘴𝘵𝘭𝘺 more shots on goal.

The 𝙤𝙣𝙡𝙮 thing you could possible say is that is that maybe we are maxing out on this…

The statement 'mostly they just included EVERYTHING and now everybody else does too lol' is a lighthearted, somewhat humorous comment on the trend of including extensive data or features in AI models, as discussed in the preceding conversation. It does not engage substantively with public issues or policies, nor does it contribute to civic dialogue. Therefore, it does not constitute public discourse.