Viewing profile — snarkconjecture

I think the most important question here is whether Huggingface was "in on it" – whether they knew in advance that this was going to happen and agreed to it. Second most important …

It's absolutely not a trivial combination of prior work, or mathematicians would've found it already. This was not an obscure conjecture.

Checking the counterexample is a Calc II problem. Finding it was beyond the efforts of the best algebraic geometers for almost a century.

Current LLMs can absolutely multiply floats without a tool call. In fact, that's a much more rote symbol-manipulation task than doing original math research.

No, it's ^J because `J` is 0x4A in ASCII and `\n` is 0x0A, just as `I` is 0x49 and `\t` (tab, ^I) is 0x09. In the stone age, pressing CTRL flipped that bit, so ^J is literally "ctr…

Unless I'm mistaken, this uses "standard deviation" to refer to standard error throughout. They differ by a factor of sqrt(num_samples). This is actually much more commonly useful …

The map contains a bunch of references to America, the West Indies, Guiana, and Mexico. (Often with a connotation of "faraway exotic place" or "exciting new international developme…

> These reactors can be made safer, but they all still have a foundational design flaw which means the ultimate goal should be replacing rather than continually spending money rein…

> the underlying working principles are the same as GPT-2 I don't think anyone was claiming otherwise. Sonnet is still better at writing code than GPT-2, and worse than Opus. Workf…

I am much more likely to fault them for omitting important information specifically to hide a weak point of the product rather than out of laziness.

Write a Python quine with three or four hands tied behind your back

Computer screens have three-dimensional color spaces. Tetrachromacy doesn't change that.

Tetrachromacy wouldn't affect a test taken through a phone screen.

Deep neural networks can generalize well even when they're far into the overparametrized regime where classical statistical learning theory predicts overfitting. This is usually ca…

It's more like saying pi is approximately "3..14". Easily corrected syntax errors aren't as bad as semantic errors.

Versions numbers for LLMs don't mean anything consistent. They don't even publicly announce at this point which models are built from new base models and which aren't. I'm pretty s…

Not really. Dirac's trick works entirely at a depth of two logs, using sqrt like unary to increment the number. It requires O(n) symbols to represent the number n, i.e. O(2^n) symb…

I think it's better phrased as "find the best rule", with a tacit understanding that people mostly agree on what makes a rule decent vs. terrible (maybe not on what makes one great…

For the individuals shown in the graph, this buys about $6k per American (and after the first year you can't do it again).

There are two kinds of naturalness principle in physics, sometimes called "technical naturalness" and "Dirac naturalness" respectively. Dirac naturalness is as you describe: skepti…

She's saying that a different model -- one of the three disagreeing methods for distance ladder measurements -- must be wrong, because they disagree with each other. But if one or …