The author uses these ([1][2]) diagrams to argue that more compute has diminishing returns. But the 'diminishing returns' are on the accuracy of correctly picking the single right category for a photo out of one thousand. Photos may simply not carry enough information to be able to meaningfully distinguish between them at that level of accuracy; existing models already exceeded humans' ability at top-5 accuracy in 2015 [3]. It wouldn't be surprising if SOTA models exceeded humans at top-1 already.
It's possible that the human baselines were bored and so performed sub-optimally when picking between the 1K classes. But the argument has now become a subtler one, much less clear cut.
As an example of categories that may be difficult to distinguish between, do you feel confident that you can reliably distinguish between the Norwich terrier [4] and the Norfolk terrier [5]? These are two separate categories in ImageNet1k.
Not being able to know things with certainty is one thing, but I don't think your latter argument about truth being literally different in different times is what the parent is getting at.
We can get really, really certain about some things. If a superforecaster (or anyone for that matter) gives 95% certainty on an event occuring and he or she has frequently made similar predictions with 95% accuracy, the event is likely to occur. It's still uncertain, but not by very much.
"Donald Trump is president". It's uncertain, certainly, but I still feel comfortable calling it a fact. Are you talking about misuse of 'facts are unquestionable' to apply to statements that aren't as clear?
https://openai.com/blog/ai-and-compute/
The amount of compute required for Imagenet classification has been exponentially decreasing:
https://openai.com/blog/ai-and-efficiency/