3 ms·
Proofs are of course the gold standard. But even if others discovered the rule, their contribution should not be held to be trivial. A modern analogy: today, p
by sn41 3y ago
Proofs are of course the gold standard. But even if others discovered the rule, their contribution should not be held to be trivial.
A modern analogy: today, people are very happy to use LLMs and Transformers without anyone "proving" that they work. Right now, philosophically, they are at the level of empirical observations (perhaps not even that). Does that mean that today's AI researchers should get no credit when at a later date? I am not sure. Empirical discovery of a rule is also no trivial thing.
- Turing_Machine 3y agoPointing out that it's not a theorem doesn't mean it's been trivialized. If there's no proof, it's not a theorem, by definition. That doesn't make empirical observations worthless -- far from it!
- saithound 3y agoYes, when people in the 4000s write clickbait holonet articles about how the Master Theorem of Neural Network Scaling was not actually invented by the Muskovites of Mars, but was found written on an Ancient American tablet containing the works of Hoffmann et al., I hope somebody will offer substantial corrections. That's not in any sense a value judgment of the empirical work done at DeepMind. Nor does it stop anybody from writing a better article which explains that the Babylonians (and many others) used the empirical observations underlying the theorem, while explaining that this did not constitute mathematical proof.
- outrun86 3y agoFound on an ancient tablet as the work of Schmidhuber et al. but ok