[voces]

Pretty meta, but I thought it was relevant here. We are familiar with Brandolini's law:

> The amount of energy needed to refute bullshit is an order of magnitude bigger than to produce it.

This can be illustrated with math or logic statements. To refute the program "1 + 1 = 3" you need to, at minimum, state "1 + 1 != 3", and such a program is always lengthier. A fuller refutation could be "1 + 1 != 3, 1 + 1 = 2", more than twice as long as the bullshit statement.

What's happening here is sort of an inverse Brandolini's law: 35 world-class computer scientists use a massive amount of programming and compute to come up with a new language model trained on massive amounts of data. The trained weights don't even fit into memory. Impressive NLP progress.

Then Gary Marcus comes around and states "Not AGI!". Not one of the computer scientists stated that they delivered AGI. But some tech journalists did. So OpenAI is guilty by association. Even though Altman came out to temper the hype and expectations. That's like proving the Poincaré conjecture, and someone dissing your research, because "1 + 1 != 3".