canjobear

It's a pdf, you can zoom in as much as you want? https://www.usenix.org/system/files/1311_05-08_mickens.pdf

Enjoy this audio version with meter, rhythm, and fully reconstructed pronunciation including pitch accent and digammas! https://hypotactic.com/my-reading-of-homer-work-in-progress/

It's good for people to change their mind in response to data.

Does their activity support healthiness, or does their healthiness support activity?

Seiken Densetsu!

Big OpenAI releases usually seem to come with some kind of baked-in controversy, usually around keeping something secret. For example they originally refused to release the weights to GPT-2 because it was "too dangerous" (lol), generating a lot of buzz, right before they went for-profit. For GPT-3 they never released the weights. I wonder if it's an intentional pattern to generate press and plant the idea that their models are scarily powerful.

Going back 3 years, it feels like incredible progress. Going back 1 year, it feels like pretty much the same limitations.

First shot, I gave it a medium-difficulty math problem, something I actually wanted the answer to (derive the KL divergence between two Laplace distributions). It thought for a long time, and still got it wrong, producing a plausible but wrong answer. After some prodding, it revised itself and then got it wrong again. I still feel that I can't rely on these systems.

It's true that Maya are still around today. But there is a legitimate big mystery of why their large-scale civilization collapsed, well before the arrival of Europeans. I think what you heard in school was some kind of telephone-game-like corruption of the legitimate question.

https://en.wikipedia.org/wiki/Classic_Maya_collapse

Or someone from South America reached them. That finding got its due attention a few years ago https://www.nature.com/articles/s41586-020-2487-2

Many have claimed to translate parts of it. What makes you find this one particularly convincing?

The Taiwan Strait is still pretty important.

> the facts about physical phenomenon that can be learned through language are different than the physical experience of physical phenomena themselves

Whether she learned through language is irrelevant. Before seeing red, she might have learned about it by consulting extensive charts of the brain, or (black-and-white) videos of timecourses of brain activation. Maybe she even got to run computer simulations of her own brain response.

By construction Mary already knew everything physical about the color red, so the new knowledge she gained through consciously experiencing it is not knowledge about anything physical.

The nonphysicality of consciousness isn't taken as axiomatic, there are arguments for it. Consider the argument from knowledge, summarized in TFA:

> A related argument known as the “knowledge argument” was famously put forward by Frank Jackson. Imagine Mary, a scientist of the future who, for whatever reason, has spent her entire life in a black and white room, never having experiences of colors. She has, nevertheless, through her studies come to learn all the physical facts there are to know about the physics and physiology of color perception. For example, she knows down to the last detail what is going on in the surface of a red apple, and in the eyes and nervous system, when someone sees the apple. Suppose she leaves the room and finally comes to learn for herself what it is like to see red. In other words, she comes for the first time to have the qualia associated with the conscious experience of seeing a red apple. Surely she has learned something new. But since, by hypothesis, she already knew all the physical facts there were to know about the situation, her new knowledge of the qualia in question must be knowledge of something over and above the physical facts.

In the thought experiment, when Mary sees red for the first time, it seems that she gains new knowledge that she did not have before even though she knew everything there is to know about the properties of matter involved. You have to either accept that she learned something nonphysical, or deny that she learned anything. The version of this that I learned first was "what is it like to be a bat?" https://en.wikipedia.org/wiki/What_Is_It_Like_to_Be_a_Bat%3F

Another one that I can recall offhand, from Descartes: Physical things have extension in space. One way to test whether something has extension in space is whether you can cut it in half. But your consciousness can't be cut in half, you can't even conceive of what this would mean. So it's not physical.

Make of the arguments what you may. I wrote term papers against both of the above in undergrad. But it's definitely something people have thought carefully about, not just an ideological assertion that consciousness Must Be Different.

For some reason cruises bring out a lot of judgment in people. But at the end of the day, some people like cruises, others don't. Some people like exploring cultures, other people enjoy entertainment at sea. Some people like roughing it in an exotic foreign place, others prefer luxury hotels. The same person's preferences might change over time. Why do you think your preferences are better than someone else's?

This kind of judgmental attitude is the thing that's not worthy of respect, imo.

Leslie Valiant was on Sean Carroll's podcast to talk about the same topic. https://www.youtube.com/watch?v=FHW-nBIZc2g

Pig butchering scams are characterized by some number of the following:

- Finding marks by sending out text messages which have the appearance of being a normal message sent to the wrong number, and which invite a response.

- Building a long-term relationship with the mark over text messages.

- Eventually convincing the mark to invest in a fake crypto exchange.

- The fake crypto exchange delivers small wins for a while

- Eventually convincing the mark to make a very large investment.

- Then the money disappears.

- The scam is run by Chinese operators in an illegal call center in Myanmar using human-trafficked labor.

This obviously has characteristics that distinguish it from a generic scam.

Example of a scam that is not pig butchering:

- I got a text message offering me $XYZ/day to drive around town with a Colgate toothpaste ad on my car

- If I had accepted the offer, they would have asked me to send in a downpayment for the ad materials, then they would have vanished

The line is the reason you don’t lose a day now. When you circumnavigate the world westward, the earth has rotated one less time for you than for everyone else, so it appears you have lost a day. By artificially switching the day at the date line you avoid this kind of slippage.

There are prestigious open-access journals that are curated in exactly the same way Elsevier ones are. Curation is completely orthogonal to whether a paper is owned by the publisher cartels.

“Talk to Transformer” GPT-2 was very funny, then GPT-3 was not funny at all. We blew through the uncanny valley in a couple of years.

Do you have examples in mind of successful foreign manipulation of US politics?

It’s true that the Transformer architecture was developed for seq2seq MT, but you can get similar performance with Mamba or RWKV or other new non-Transformer architectures. It seems that what is important is having a strong general sequence-learning architecture plus tons of data.

> The GPT architecture, using transformers to do iterated predictive text, is a modern version of the Markov bot.

The Markov nature only matters if the text falls outside the context window.

> Perhaps surprisingly so, until you step back, look at the training data, and look at the information flow: the conditional probability of the next token isn't mostly coming from the source text.

I’m not sure what you’re getting at here. If it’s that you can predict the next token in many cases without looking at the source language, then that’s also true for traditional encoder-decoder architectures, so it’s not a problem unique to prompting. Or are you getting at problems arising from teacher-forcing?

Basically the question was how an LM could possibly help translation, and the answer is that it gives you a strong prior for the decoder. That’s also the basic idea in the theoretical UMT paper: you are trying to find a function from source to target language that produces a sensible distribution as defined by an LM.

“Throw an LM at it” is the only approach that has ever produced human level machine translation.

For theory on how a strong target-language-side LM can improve translation, even in the extreme scenario where no parallel “texts” are available, https://proceedings.neurips.cc/paper_files/paper/2023/file/7...

What do you think is being obfuscated? It’s normal to give money so you can name stuff.

The Taco Bell professorship is an endowed chair, not something you apply for. It’s paid for by Taco Bell as a gift to the university, then the university awards it as part of a recruitment (or retention) process for senior professors. It means the professor gets a nice salary and research stipend that is independent from the rest of the school, and can live the life of research and thinking.

I don’t know much about Def Con but the drama coming out of this conference is really entertaining.

I believe this is called a “junk drawer”

I’m sitting in an electric car in a large Chinese city right now. Looking around I count 7/10 cars are electric, easily recognizable by green license plates, all from brands I’ve never seen in the US. The electric cars are cheap and I don’t see any evidence that they are poor quality. It feels like there’s absolutely something transformative happening here which the US is missing out on.

How good is the accuracy of paleological methods? Have they been tested in scenarios where a ground truth later became available?

I doubt this could possibly have been a controlled experiment.