ChadNauseam

> Example: memorize the major scale as a series of whole and half steps but we won’t provide you with any context or motivation for why this is so.

I recommend "What Makes Music Sound Good?" by Prof. Dmitri Tymoczko: https://dmitri.mycpanel.princeton.edu/files/pdfs/MUS105hando...

There is a part near the beginning where he derives the diatonic scale from first principles. (There is a sequel as well, and a book if you prefer to read it in book form)

I don't think 100% accuracy is logically possible. Because it's entirely possible that someone would just naturally write the exact same thing as an LLM would write. And after the fact there is no way to distinguish the two. But pangram does have an extremely low false positive rate, which I think does make it useful for detecting cheating students. Assuming the base rate of cheating students is 1%, and assuming pangram has a false positive rate of 1 in 10,000 and a true positive rate of 7,000 in 10,000, that means ~98% of students flagged by pangram actually cheated. Combined with a teacher's familiarity with that student's previous work, which should rule out many more false positives, it should be a very useful tool.

> And then you'd ask it about an area you're knowledgeable in and realise it routinely makes stupid mistakes.

This honestly doesn’t happen to me much anymore. In what areas do you find LLMs routinely make stupid mistakes?

> In my view, intelligence includes an ability to learn and adapt to never-before-seen situations.

This is exactly what ARC AGI tests

> And then general intelligence is an ability to apply that across a wide variety of domains.

My experience with Fable is that it can certainly apply that in a wide variety of domains

> However, last I checked, a seemingly very intelligent LLM still struggles to play Chess at a basic level

People also struggle to play Chess at a basic level. They only succeed by studying the game for a long time. I will concede that humans can do this and LLMs generally cannot.

Actually one of the worst and most insane file formats of all time. I partially reverse engineered it a while ago. My parser is super janky and I can't share it, but I pasted the code into claude and had it summarize it: https://gist.github.com/anchpop/a14325cc451b04a5bf78c476ac20...

PhD programs in technical fields usually give you a stipend

my personal favorite, although css not html, is “field-sizing: content”. Finally with iOS 26 we can have input fields that expand to fit the content

I'm reminded of this tweet by roon:

> it is a literal and useful description of anthropic that it is an organization that loves and worships claude, is run in significant part by claude, and studies and builds claude. this phenomenon is also partially true of other labs like openai but currently exists in its most potent form there. i am not certain but I would guess claude will have a role in running cultural screens on new applicants, will help write performance reviews, and so will begin to select and shape the people around it.

https://x.com/tszzl/status/2051045196260167790

> One can explain it as the computer checked a bunch of cases and all maps reduce to one of these cases. It doesn’t take an expert to state this.

Hmm, doesn't it take an expert to explain why those cases are exhaustive, and why the code that checked them is correct?

Tangentially, I'm not a mathematician but I wonder if one "opaque" proof that is too complicated for anyone to understand, but that we know is correct via formal verification, might end up being built on with "transparent" human-understandable proofs. For example, it's my understanding that there are many conjectures that have been proven true conditional on the riemann hypothesis being true. In that case, an opaque proof of the riemann hypothesis would enable those conjectures to be known and built upon

That's interesting, thanks for sharing. Maybe the classifiers are really sensitive to anything that has to do with memory safety.

I left both comments 12 hours ago. The first reply to my other comment was 11 hours ago.

> if you ask something too obvious or direct it will stop you other than that, it work flawlessly

What on earth are you asking it?

Yeah I feel like I'm living in a different dimension than these people. I wonder what they're working on. I've literally never had it refuse everything and I max out my 20x plan every week

> This assertion is "oddly" similar to the GPT answer "neural network inference determinism" produced

My statement and GPT's statement are both correct answers to the same question, so I think it makes sense that they would be similar. Are you accusing me of having paraphrased an LLM in writing my answer? I did not, I just remembered having read Thinky's post on the subject [0], which GPT has probably read also.

> This holds for a vanishingly small set of conditions, none of which include randomness, nor when context and transformers are involved, let alone underlying model evolution (thus making model use over time non-deterministic).

There are plenty of ways to introduce nondeterminism into any system. By your standards, I doubt you could point to a single deterministic system in the world. print("hello, world") is only deterministic if your CPU is properly shielded from cosmic rays and your OS isn't out of memory etc. There are some inherently nondeterministic processes, like the stochastic methods used to train models or the random sampling used at inference time if you have temperature!=0, but inference under greedy decoding is conceptually deterministic.

[0]: https://thinkingmachines.ai/blog/defeating-nondeterminism-in...

Zcash and many other cryptocurrency-related projects. The amount of cryptographers working on zero-knowledge proofs probably multiplied 10x thanks to cryptocurrencies.

It's a little bit funny to me. I can understand a response of "Hmm, while your application of cryptography may have some legitimate uses, in practice it seems like it's largely used for crime and scams." But I wonder what they thought the primary application of cryptography was going to me! Privacy-preserving technologies that allow you to hide from the government are necessarily going to be useful to criminals. But that's the price of freedom, no?

(I guess to be fair, some cryptographers might be interested in things like privacy-preserving voting, which wouldn't ever be useful to criminals.)

There's nothing inherently biased or intentionally wrong with greedy decoding. Why would there be? Let's say you're trying to predict what an expert doctor would say to a patient with cancer. You think there's a 99% chance the doctor would say "start chemotherapy" and a 1% chance the doctor would say "don't worry about it". If you have to pick one to output, you'll pick "start chemotherapy". It would be crazy to say the best thing to do is to roll a 100-sided die to determine your answer. A language model is literally in that exact situation.

Now that you mention it, I think "statistical" might be a good word choice as it probably closely matches what people mean when they say an LLM is nondeterministic. However, on a technical level, neural network inference truly is inherently deterministic. The only nondeterminism in LLMs is parallelism-dependency in the way floating point operations are carried out and in the decoding strategy

I feel like I'm fighting a losing battle but I don't see why so many people consider LLMs innately non-deterministic, an LLM running on a CPU with greedy decoding would be 100% deterministic and still have the intelligence we associate with LLMs

The technology is still in its infancy unfortunately, so there's no way the web could have been based on it, but I think content-addressing is the long-term play. If I click a link, there are some cases where I want the server to respond with a fresh response just for me (e.g. a website showing the weather). But often I just want whatever content was linked to (e.g. a webpage explaining a math content). In the latter case, it would be nice if the link had a hash of the content in it, and 3rd parties could host copies to keep the link working even if the original operator stopped existing.

Is this a linux issue? My Macbook has 16gb ram and running out has literally never been an issue for me. I'm using all of it + 10gb of swap and my computer feels snappy and fast. Chrome with 100+ tabs, discord, slack, codex, chatgpt, all running simultaneously with no issues. Sure, all other things being equal less RAM usage is better, but I can understand why reducing it isn't a priority for them.

I've been pretty impressed with Laguna. I downloaded their coding agent and have used it for a task here and there (the larger variant). Obviously it's nothing like a frontier LLM, but it surprised me with how good it was. And I think the model personality and way it talks is pretty pleasant

Is what still an issue? This is from the article

> Using a novel cross-industry design and the 1999–2003 expansion of the H-1B visa cap for identification, we find that H-1B exposure raised incomes for natives and pre-existing immigrants, with gains concentrated in non-STEM occupations. Income gains propagate forward through supply chains to downstream industries but not backward to upstream industries, consistent with a productivity shock rather than a labor supply shock.

I've always thought it was unfair that the 9 in 9% uptime doesn't count.

Someone who would not otherwise be able to asks AI for simple instructions for how to make a deadly virus, which then spreads rapidly and kills most of those it infects

> enabling those who believe themselves to be good is more important than reducing wanton suffering

I'm genuinely not sure if you're referring to my comment or the one I responded to haha

I thought this might have something to do with sentencepiece's unigram (which uses viterbi). But they seem to be totally different domains. What an amazing algorithm, to show up in so many different places.

It's an interesting time when the bar has moved to "It’s not like diplomacy and war will end". But that is on the table. Nuclear weapons caused the end of wars as we know them. Great powers no longer directly attack each other. And AI is much more powerful than nuclear weapons, with an equal capacity for damage and a much greater capacity for good.

By the way, I donate monthly to GiveWell and the shrimp welfare project. Do you donate monthly to starving people? Most people don't, and the reason is simple: at the end of the day, most people just don't care that much about helping a starving person far away. They also don't care much about things like that, animals in factory farms, or earthquakes that kill hundreds of thousands. But ideally, we could find a way to empower people such that the minority who do care can make a big difference.

> Likely 10-1000x cheaper to implement than fully autonomous cars.

I think self driving buses would be awesome. I can imagine having smaller ones that come more regularly too. Add some amenities like a tracker that tells you exactly when to leave to catch the bus, and exactly when you'll reach your destination, and it would probably be an awesome UX.

But I don't know if I agree that it'd be 10-1000x cheaper to implement than fully autonomous cars. And the overhead of a car is huge, you need one driver for a maximum of 4 passengers and probably a mode of 1 passenger. The overhead of a bus is smaller, because one driver can carry 30 or 60 people. Arguably, you'd need less of them because busses carry more people, but for some reason the unit economics often doesn't work out like that IRL. It'd make more sense for a neighborhood to have just one xbox that everyone shares when they need it, rather than everyone having their own xbox that's idle 95% of the time, but that's not how it went. (Same for any other computing device.)

> if the were competent they could spend 6 month and few 100 million and build absolutely amazing bus transit system

$346 million gets you utility improvements and 2 miles of bus lane: https://en.wikipedia.org/wiki/Van_Ness_Bus_Rapid_Transit

They absolutely will. I'm a very social person but when I'm going home from a date with my girlfriend I always want a Waymo just because I don't want a 3rd person involved. And sometimes I'm having a bad day and don't want to have to make small talk. And it's nice that I can talk on the phone without disturbing someone.