RyEgswuCsn

> legacy media (propaganda machines)

I'd say that qualification requires some upgrade. What's worse than an obvious propaganda machine?

A machine.

As for the 1A --- I didn't mean to fault it; it is not its fault.

I think they do care. People thinking the U.S. government doesn't care may as well be an indication that their efforts on propaganda are highly effective and successful.

> I'm not sure that the 1A can scale properly for modernity, though.

I agree. I think the modern-day challenge is not that one cannot speak freely, but that their voice cannot compete with institution-sponsored propaganda machines --- after all, humans have limited attention. The situation is arguably worse because it suppresses free speech without being appear to suppress it.

> Propaganda is widely used on social media. Why focus on TikTok?

Because TikTok is a "rogue" social media (from the U.S.'s perspective, of course)?

> Other countries have laws and agencies to protect against propaganda; the US doesn't bother.

I am not sure I follow your point here; could you elaborate?

I don't think it is just about TikTok being capable of spying on foreign users; my opinion is that it is more about TikTok's propaganda potential, i.e. how it can be used to manipulate narratives in the most subtle way against U.S.'s national interest.

Imagine the case where the U.S. government would like the (western) world to form certain sentiments on the current regional conflicts that are aligned with its own interests, and then imagine how TikTok's recommendations could be deliberately designed to compromise such alignment...

Well, I am afraid it will be either you doing it (with the help of AI), or your client/boss doing it using AI without you.

That is very interesting. Skimming through the thesis, I find it fascinating introductory material to Nix. And you are correct, Nix started as a deployment tool. Although I'd argue in an era (1978!) when software are mostly distributed as source, reliable deployment largely implies reliable building from source.

My understanding is that Nix was initially invented to solved the "producing builds with a fully specified dependencies graph" problem; people then figured out that it can dub as a package manager to support a better developer experience.

Docker, on the other hand, was introduced to solve the deployment problem --- it provides light-weight isolation so that you get boggled in resolving dependency conflicts. Of course, people later also use it drive development environments, which I find quite sensible.

Found it: https://github.com/spack/spack

Perhaps it is just me. But after so many attempts of trying to getting started with Nix, it is still not clear to me how to *author* Nix packages. My mental model about Nix is that it resembles make, but for the whole dependency graph. I kind of hoped that the Nix language is more sane. I think there is a project of similar goals but the script is in Python? I remember finding it once, but couldn't recall its name.

But that is not the LLM learning to produce valid JSON —— as some other commenters mentioned, you can get valid json without the LLM.

Sure it could be useful, but not really impressive.

How does the LLM know what valid JSON tokens are?

What if the training data contains malformed JSON? There ought to be a non-zero chance of the LLM producing invalid JSON, no?

"Never stop fighting for the control of the agenda, for all your enemies are doing the same."

--- Chairman Mao

Do you support server pushes, e.g. periodic updates initiated by the server to update client UIs?

I am no expert on this topic, but I feel this is similar to how ferrofluid forms "spikes" when placed near a magnet. Are they related?

I knew it!

Yeah, I really meant coal-powered steam engines.

Are they really the same though? I mean we are talking about room temperature, not the"(relatively) high temperature", a term that is arguably misleading to a layperson.

I'd imagine most people who are informed of the difference would recognize that cuprate SC is nowhere near as useful.

Petrol is just "better" coal. Except that you can't make flight with steam engines.

When room-temperature SC becomes possible, previously unthinkable applications suddenly become viable.

I kind of hoped that new contenders to LaTeX could ditch the state-based sections (e.g. `\section{Introduction}`) and switch to strictly scoped sections that are properly closed (i.e. section trees).

I was thinking more of the entropy of the distribution over the distribution sort of thing --- one may say a model is uncertain about its statements if all its potential variants are equally probable and vice versa.

Many classic statistical modelling techniques have ways to produce some measure of confidence for their predictions; perhaps LLMs could incorporate that as well, e.g. assign probabilities(/perplexity?) to each of the tokens they generate.

> LLMs are great at natural language comprehension, but also have a lot of neurons dedicated to factoids. Using neurons that way is really inefficient, can we split the "language" capability from the "knowledge" capability and have the former just look things up in a database?

I think the beauty of LLMs is exactly that all we need to do is to feed them raw text --- the hope, I guess, had been that the models will be able to develop human-like insights by learning to understand and "speak" languages on its own. Introducing "feature engineering" (e.g. the distinction as you suggested) would defeat that goal.

I think LLMs need to be taught to say "I don't know"/"I am not sure" or something to that effect. Another approach might be to introduce an adversarial "censor" model to guard against hallucination (or inappropriate answers).

It sort of does. RO produces relative large amount of waste water that are harder to deal with (e.g. storage, disposal). By spending more energy, you can turn the waste water into salt solids, which can be easier to dispose of.

Well, there is always the option of C/C++...

I am not saying that the redistribution of risk is not useful —— it certainly is, and I agree with what you said. But let us suppose we would like to reverse climate change at a global scale in a short time without further damaging the environment, right now; I don’t see how it would be possible with our current technologies, even if every possible risk redistribution options are exhausted.

Except that is not exactly "productive", isn't it? After all, risk was not eliminated, only redistributed. Productive output, e.g., would be something that reduces the chance of your house catching fire.

How do people feel about having some of the open-source LLMs offloading some of the moderation burdens? I understand it sounds rather dystopian, but I also feel if the model is developed transparently, it may offer a solution to the moderation problem.

> refactoring tools

Exactly.

Then the problem isn't IDEs cripple productivity, it is simply that one may not be operating at a scale that benefit from using them.

There are many decent open source IDEs (e.g. NetBeans and Eclipse for Java).

I don't know. Can you reliably rename variables/classes in a language that is not designed to facilitate IDEs/toolings?