canjobear

There was no announcement. This is just a conjecture based on Kremlinology (Pyongyangology?) by a South Korean government agency.

There's already a huge amount of AI generated music on Youtube.

I wonder about the accuracy of critical text methods like the ones that have been used to putatively reconstruct the Q document and to argue about authorship and dates. Have these methods ever been validated against a ground truth that the arguers didn't know about beforehand? Like, have we ever philologically reconstructed a text from other texts, and then found exactly that text buried somewhere? Or even something close to it?

In the case of Q, you could argue that the Gospel of Thomas validates that there were texts of that kind (sayings gospels) floating around, but Thomas doesn't match the content of Q.

Outside biblical scholarship, another area where people have tried to reconstruct what is going on in ancient texts is the Chinese classics, especially the really cryptic ones like the Yijing. But whenever some actual ancient manuscript gets dug out of an old grave or a bog, it seems like it just brings up more questions and complications, instead of validating anyone's theories.

Compare to the philology methods that people use to reconstruct ancient languages. These have been validated pretty well. For example in the 19th century linguists were able to deduce that the Proto-Indo-European language must have had guttural consonants not found in any extant language, and then later when the Hittite language was decoded, the guttural consonants were right there. The theory was validated on held-out data. Has this ever happened for critical methods for discerning authorship and sources and missing texts?

He's criticizing other people's attitudes there, not stating his own.

I read a good way down thinking this was some kind of highly metaphorical blog post about the programming language Julia

Does "fake" here mean "real"?

The effect replicates in languages with other writing systems.

“who art in heaven” is a grammatical relative clause because the subject of the verb is the relative pronoun “who” which is second person in that context. You can still get this kind of thing in modern English, for example “I, who am a farmer, will be happy” is grammatical because the relative pronoun “who” is first person there. That doesn’t mean it would be grammatical to say “*A farmer am happy” and it wouldn’t have worked with art either.

Conceivably it’s grammatical if Henry is vocative and the pronoun is dropped colloquially, like “Who art [thou], O Henry?” but it’s a stretch.

"Who art Henry?" was never grammatical English. "Art" was the second person singular present form of "to be" and it was already archaic by the 17th century. "Who is Henry?" would be fine.

Diet Coke is really the sugar free version of New Coke, now discontinued. It wasn’t meant to taste the same as Coca-Cola Classic.

> the bar for the test gets raised and eventually humans won’t pass the test themselves.

At this point LLMs usually beat humans at the Turing Test! People are more likely to pick the LLM as the human, rather than the human. https://arxiv.org/abs/2503.23674

The chart is questions posted. Almost all my SO usage was looking at answers already posted. I only posted questions when I couldn’t find anything. I wonder if the pre-2023 decline is just because more and more of the questions people had were already covered.

I thought it was Guangxi man...

I think I understand your argument (you could define "the smallest 'undefinable' number" and now it has a definition) but I still don't see how it overcomes the fact that there are a countable number of strings and an uncountable number of reals. Can you exhibit a bijection between finite-length strings and the real numbers? It seems like any purported such function could be diagonalized.

No, you can establish a bijection between strings and natural numbers, very easily.

I think you're overthinking it. Define a "number definition system" to be any (maybe partial) mapping from finite-length strings on a finite alphabet to numbers. The string that maps to a number is the number's definition in the system. Then for any number definition system, almost all real numbers have no definition.

This was the (juvenile) nerd take back then https://maddox.xmission.com/c.cgi?u=iphone/

(Nokia E70 not N95, but still)

Chaitin's constant is definable but not computable.

> 1: Almost all numbers are transcendental.

Even crazier than that: almost all numbers cannot be defined with any finite expression.

Upper left corner of the Enterprise-D bridge, you can see the door to the bathroom.

Are there particular successes that you have in mind? Because what comes to my mind is stuff like https://en.wikipedia.org/wiki/Deportation_of_the_Crimean_Tat...

I’m pretty sure the idea that things can be good in some ways and bad in other ways came way before Kant.

It’s more like if the supervillain had to write one word of his chain of thought, then go away and forget what he was thinking, then come back and write one more word based on what he had written so far, repeating the process until the whole chain of thought is written out. Each token is generated conditional only on the previous tokens.

They represented consonants in the Egyptian language, but by convention we now read them as vowels. 𓄿 was probably a glottal stop, 𓇋 was probably like the y in yet. https://en.wikipedia.org/wiki/Egyptian_language#Egyptologica...

It also seems possible that they were sometimes used to stand for vowels even in real Egyptian phonology, in the same way that certain consonant signs are used in Hebrew and Arabic https://en.wikipedia.org/wiki/Mater_lectionis

> but DL is proven to be able to find pattern even in random noise, so how can you be sure this is not just overfitting to artefact?

You test your DL decoder on held-out data. This is the common practice.

> The same with knowledge and understanding, you can organize material so they don’t end up in unproductive rabbit holes - but they have to work out their understanding on their own.

Problem sets with feedback.

This is right in terms of the rigorous statistical sense of “Markov model”. But in practice in the world of NLP and chatbots through the 90s and 2000s, “Markov model” was usually used to refer to Markov chains (ie you only condition on the previous k words). Hence the term “Hidden Markov Model” to refer to what you’re calling a Markov model.

What's the relationship with the Platonic Representation Hypothesis?

They did a Bonferroni correction to the p values, which is supposed to compensate for this kind of thing.