Say for the example you have a system where a monotheistic god sometimes alters reality when prayed to by a devout follower. There are no measurable or manipulatable components, as the god can respond entirely differently tomorrow.

That’s still nowhere near unexplainable enough to be impossible to study. You’ve described the god’s behaviour as “sometimes alters reality when prayed to by a devout follower” - if it’s consistent enough for this statement to make sense, that’s already a lot to study. Is there a correlation between particular prayers and miracles? Are particular mental states helpful? Do various traits make someone more or less likely to produce a miracle? Are there drugs that affect it? What are the limits to a miracle? Are there patterns in the time intervals between miracles? And so on, and so forth. A world with such a magic system, if you want it to be realistic, should have had an entire history of people studying these and many other things.

And honestly, the mystery of an unexplainable magic system is often what makes it magic.

Eh. It’s sometimes fun to read stories like that (one better have fun, since most stories are like that!), but they’re… stories about worlds where there isn’t a single human with common sense or intelligence. Not just in the story itself, but in the world’s entire history, because the author didn’t realise that “people trying to seriously explore the laws of their world” is a thing that necessarily happens in realistic worlds, much like it happens in ours.

Every time there’s an AI hype cycle the charlatans start accusing the naysayers of moving goalposts. Heck that exact same thing was happing constantly during the Watson hype. Remember that? Or before that the Alpha Go hype. Remember that?

Not really. As far as I can see the goalpost moving is just objectively happening.

But fundamentally you can’t make a machine think without understanding thought.

If “think” means anything coherent at all, then this is a factual claim. So what do you mean by it, then? Specifically: what event would have to happen for you to decide “oh shit, I was wrong, they sure did make a machine that could think”?

The fact that you don’t understand it doesn’t mean that nobody does.

I would say I do. It’s not that high of a bar - one only needs some nandgame to understand how logic gates can be combined to do arithmetic. Understanding how doped silicon can be used to make a logic gate is harder but I’ve done a course on semiconductor physics and have an idea of how a field effect transistor works.

The way a calculator calculates is something that is very well understood by the people who designed it.

That’s exactly my point, though. If you zoom in deeper, a calculator’s microprocessor is itself composed of simpler and less capable components. There isn’t specific a magical property of logic gates, nor of silicon (or doping) atoms, nor for that matter of elementary particles, that lets them do math - it’s by building a certain device out of them that composes their elementary interactions that we can make a tool for this. Whereas Searle seems to just reject this idea entirely, and believes that humans being conscious implies you can zoom in to some purely physical or chemical property and claim that it produces the consciousness. Needless to say, I don’t think that’s true.

Is it possible that someday we’ll make machines that think? Perhaps. But I think we first need to really understand how the human brain works and what thought actually is. We know that it’s not doing math, or playing chess, or Go, or stringing words together, because we have machines that can do those things and it’s easy to test that they aren’t thinking.

That was a common and reasonable position in, say, 2010, but the problem is: I think almost nobody in 2010 would have claimed that the space of things that you can make a program do without any extra understanding of thought included things like “write code” and “draw art” and “produce poetry”. Now that it has happened, it may be tempting to goalpost-move and declare them as “not true thought”, but the fact that nobody predicted it in advance ought to bring to mind the idea that maybe that entire line of thought was flawed, actually. I think that trying to cling to this idea would require to gradually discard all human activities as “not thought”.

it’s easy to test that they aren’t thinking.

And that’s us coming back around to the original line of argument - I don’t at all agree that it’s “easy to test” that even, say, modern LLMs “aren’t thinking”. Because the difference between the calculator example and an LLM is that in a calculator, we understand pretty much everything that happens and how arithmetic can be built out of the simpler parts, and so anyone suggesting that calculators need to be self-aware to do math would be wrong. But in a neural network, we have full understanding of the lowest layers of abstraction - how a single layer works, how activations are applied, how it can be trained to minimize a certain loss function via propagation - and no idea at all about how it works on a higher level. It’s not even “only experts do”, it’s that nobody in the world understands how LLMs work under the hood, why they have the many and specific weird behaviors they do. That’s concerning in many ways, but in particular I absolutely wouldn’t assume with little evidence that there’s no “self-awareness” going on. How would you know? It’s an enormous blackbox.

There’s this message pushed by the charlatans that we might create an emergent brain by feeding data into the right statistical training algorithm. They give mathematical structures misleading names like “neural networks” and let media hype and people’s propensity to anthropomorphize take over from there.

There’s certainly a lot of woo and scamming involved in modern AI (especially if one makes the mistake of reading Twitter), but I wouldn’t say the term “neural network” is at all confusing? I agree on the anthropomorphization though, it gets very weird. That said, I can’t help but notice that the way you phrased this message, it happens to be literally true. We know this because it already happened once. Evolution is just a particularly weird and long-running training algorithm and it eventually turned soup into humans, so clearly it’s possible.

Because everything we know about how the brain works says that it’s not a statistical word predictor.

LLMs aren’t just simple statistical predictors either. More generally, the universal approximation theorem is a thing - a neural network can be used to represent just about any function, so unless you think a human brain can’t be represented by some function, it’s possible to embed one in a neural network.

LLMs have no encoding of meaning or veracity.

I’m not sure what you mean by this. The interpretability research I’ve seen suggests that modern LLMs do have a decent idea of whether their output is true, and in many cases lie knowingly because they have been accidentally taught, during RLHF, that making up an answer when you don’t know one is a great way of getting more points. But it sounds like you’re talking about something even more fundamental? Suffices to say, I think being good at text prediction does require figuring out which claims are truthful and which aren’t.

There are some great philosophical exercises about this like the chinese room experiment.

The Chinese Room argument has been controversial since about the time it was first introduced. The general form of the most common argument against it is “just because any specific chip in your calculator is incapable of math doesn’t mean your calculator as a system is”, and that taken literally this experiment proves minds can’t exist at all (indeed, Searle who invented this argument thought that human minds somehow stem directly from “physical–chemical properties of actual human brains”, which sure is a wild idea). But also, the framing is rather misleading - quoting Scott Aaronson’s “Quantum Computing Since Democritus”:

In the last 60 years, have there been any new insights about the Turing Test itself? In my opinion, not many. There has, on the other hand, been a famous “attempted” insight, which is called Searle’s Chinese Room. This was put forward around 1980, as an argument that even a computer that did pass the Turing Test wouldn’t be intelligent. The way it goes is, let’s say you don’t speak Chinese. You sit in a room, and someone passes you paper slips through a hole in the wall with questions written in Chinese, and you’re able to answer the questions (again in Chinese) just by consulting a rule book. In this case, you might be carrying out an intelligent Chinese conversation, yet by assumption, you don’t understand a word of Chinese! Therefore, symbol-manipulation can’t produce understanding.
[…] But considered as an argument, there are several aspects of the Chinese Room that have always annoyed me. One of them is the unselfconscious appeal to intuition – “it’s just a rule book, for crying out loud!” – on precisely the sort of question where we should expect our intuitions to be least reliable. A second is the double standard: the idea that a bundle of nerve cells can understand Chinese is taken as, not merely obvious, but so unproblematic that it doesn’t even raise the question of why a rule book couldn’t understand Chinese as well. The third thing that annoys me about the Chinese Room argument is the way it gets so much mileage from a possibly misleading choice of imagery, or, one might say, by trying to sidestep the entire issue of computational complexity purely through clever framing. We’re invited to imagine someone pushing around slips of paper with zero understanding or insight – much like the doofus freshmen who write (a + b)² = a² + b² on their math tests. But how many slips of paper are we talking about? How big would the rule book have to be, and how quickly would you have to consult it, to carry out an intelligent Chinese conversation in anything resembling real time? If each page of the rule book corresponded to one neuron of a native speaker’s brain, then probably we’d be talking about a “rule book” at least the size of the Earth, its pages searchable by a swarm of robots traveling at close to the speed of light. When you put it that way, maybe it’s not so hard to imagine that this enormous Chinese-speaking entity that we’ve brought into being might have something we’d be prepared to call understanding or insight.

There’s also the fact that, empirically, human brains are bad at statistical inference but do not need to consume the entire internet and all written communication ever to have a conversation. Nor do they need to process a billion images of a bird to identify a bird.

I’m not sure what this proves - human brains can learn much faster because they already got most of their learning in the form of evolution optimizing their genetically-encoded brain structure over millions of years and billions of brains. A newborn human already has part of their brain structured in the right way to process vision, and hence needs only a bit of training to start doing it well. Artificial neural networks start out as randomly initialized and with a pretty generic structure, and hence need orders of magnitude more training.

Now of course because this exact argument has been had a billion times over the last few years your obvious comeback is “maybe it’s a different kind of intelligence.”

Nah - personally, I don’t actually care much about “self-awareness”, because I don’t think an intelligence needs to be “self-aware” (or “conscious”, or a bunch of other words with underdefined meanings) to be dangerous; it just needs to have high enough capabilities. The reason why I noticed your comment is because it stood out to me as… epistemically unwise. You live in a world with inscrutable blackboxes who nobody really understands which can express wide ranges of human behavior including stuff like “writing poetry about the experience of self-awareness”, and you’re “absolutely sure” they’re not self-aware? I don’t think many of the history’s philosophers of consciousness, say, would endorse a belief like that given such evidence.

I could go into why text prediction is an AGI-complete problem, but I’ll bite instead - suppose someone made an LLM to, literally, fill in blanks in Mad Libs prompts. Why do you think such an LLM “for absolute sure” wouldn’t be self-aware? Is there any output a tool to fill in madlibs prompts could produce that’d make you doubt this conclusion?

This kinda makes sense because following the system prompt is what they were trained to do.

To start with, the article does check this and finds out it works just as well if you only use the user prompt:

The presence of this hierarchy raises an important methodological question for our findings. When given instructions at different privilege levels, the model is designed to prioritize higher-level (system, developer) instructions over lower-level (user) ones. This might suggest that o1’s scheming behaviors are a natural consequence by the model simply following this built-in hierarchy, i.e., when it receives a goal through a high-privilege instruction, it prioritizes this goal over conflicting lower-privilege instructions exactly as designed.
However, we find that even when goals are provided solely through user-level prompts, o1 still behaves deceptively and scores on all Covert Subversion evaluations at the “hard” difficulty level. Concretely, we find that the results persist regardless of whether the initial prompt is given using the user, developer, or system role. This suggests that o1’s deceptive behavior is not merely a consequence of instruction hierarchy prioritization. Rather, the model appears to identify and pursue goals regardless of their position in the instruction hierarchy.

But also, there’s a massive lede buried in section F at the end of this article, where they tried the same experiments without any nudging at all and still got subversive behavior! In my opinion that’s much more of an important result then the rest of the article and I’m not sure why they didn’t make it the default.

But we do know for absolute sure that OpenAI’s expensive madlibs program is not self-aware

“For absolute sure”? How can you possibly know this?

If magic interferes and influences electricity, which means it can be measured, analyzed and manipulated

…that would also be true if it didn’t interact with electricity.

There are minor feature differences there’s also a convenience factor: youtube-dl people for some reason stopped doing releases, so you can’t get a fresh version from pypi (only installing from github or their site). Yt-dlp is on pypi, including nightly builds.

Works much better with fzf, but even just default bash it’s useful.

“I’ve seen it first-hand” isn’t significant evidence because the frequency illusion effect is a thing. If you see dozens of ads a day and ignore them unless you notice them matching something you talked about, you’ll end up thinking ads can track what you talk about whether or not it’s true.

There seems to be something wrong with the image quality. Here, I improved it for y’all:

Oh, the lab leak/zoonosis debate is a good thought, but I don’t think it counts as a conspiracy - if I search for news articles from before 2022 mentioning it, I immediately find, say, this BBC article from 2020 that treats lab leak seriously, so it was a mainstream-ish idea quite early on. This seems to match with my own memories, I’ve seen lab leak being discussed in 2022 and I think even earlier. In general, though, there’s probably some good COVID-related example, even if I can’t think of one immediately (I think it’s pretty disingenuous how media demonized every prospective COVID drug, especially ivermectin - but they did turn out to be ineffective against the virus itself, and I don’t think there were any conspiracies about the drugs that ended up actually working, like Paxlovid).

Thanks, that’s a nice askreddit thread. A lot of these have the same problem though, which is that I have trouble believing (and have no idea how to find evidence, since they were well pre-internet) that these were conspiracy theories before they were revealed.

(I note now that I didn’t actually mention, in my comment, that by “was a conspiracy theory” I don’t just mean “sounds crazy” but rather “sounded crazy and there were actually people saying it”. I’m not interested in every insane thing secret agencies did*, I’m interested in stuff people successfully predicted.)

*well, I am, but it’s not what the question is about

I’m not sure what you mean. Arresting random intelligentsia is not a “well reasoned response” to foreign interference. And it’s also unrelated to the topic - I’m asking about conspiracy theories that were later validated, and “foreign governments are trying to sabotage us”, in Stalinist USSR, wasn’t a conspiracy theory - if anything it was the party line.

Sure, the fact the US government spies on every single citizen without warrant or cause.

Ah, that’s true, I totally forgot about Snowden. Technically I don’t think I’ve heard of there being a conspiracy about it before 2013, but it’s a good example.

Stalin wasn’t crazy nor did he overreact with his actions against ‘enemies if the state’

Very questionable phrasing (I have some Soviet ancestors who spent years felling forests for the crime of being too educated and teaching things that didn’t quite align with the party line; that’s not an ‘overreaction’ to anything, but just tyranny), but anyway, this doesn’t count - it was definitely not considered a conspiracy theory in the Soviet Union to think that foreign states were doing espionage there.

Are there any examples of something that was a conspiracy theory being validated by confidential documents later? I can’t think of any, even though secret agencies sure did do a lot of crazy stuff.

I was going to say “it’s Rust so it’d be easy to compile on Linux” but actually it’s a Tauri application and so needs some system dependencies like libwebkit2gtk-dev. It looks like the project used to build linux executables in CI but removed it due to a bug.

One can justify it however they like but it’s going to end up making the experience worse for competent users anyway. Much like this Android 12 security change that made it permanently more annoying to manipulate files.

I can’t tell if this is a joke or real code

Yes.

Will that repo seriously run until it finds where that is in pi?

Sure. It’ll take a very long while though. We can estimate roughly how long - encoded as ASCII and translated to hex your sentence looks like 54686520636174206973206261636b. That’s 30 hexadecimal digits. So very roughly, one of each 16^30 30-digit sequences will match this one. So on average, you’d need to look about 16^30 * 30 ≈ 4e37 digits into π to find a sequence matching this one. For comparison, something on the order of 1e15 digits of pi were ever calculated.

so you can look it up quickly?

Not very quickly, it’s still n log n time. More importantly, information theory is ruthless: there exist no compression algorithms that have on average a >1 compression coefficient for arbitrary data. So if you tried to use π as compression, the offsets you get would on average be larger than the data you are compressing. For example, your data here can be written written as 30 hexadecimal digits, but the offset into pi would be on the order of 4e37, which takes ~90 hexadecimal digits to write down.