kmod

While I'm with you on being against dark patterns and unethical behavior, anti-cheat feels different in that it seems pretty honestly an effort to make the game better for players (the non cheaters, at least). I haven't seen any arguments that anti-cheat is being used in the interests of the company over the user, it's all that people don't like the technology that's required to do this. So I think it's pretty different from DRM which is pretty clearly an interest misalignment.

At the day yes no one likes this software, but people like gaming with cheaters even less. Just like how we don't like being subject to social media moderation, but we dislike spam even more. I think we can't rule out that these companies (in these instances) are trying to give us what we want even if the means are a bit unseemly.

It seems like the people who are massively over-hyping AutoGPT have never used it. It's fascinating how the entire hype cycle can be based on an idea of what's been done as opposed to what's actually been done.

As to the tool itself, I played around with it and it has some cool ideas that I think are valuable, even if it's not AGI

Just be careful, it's trained to give you answers you like, and in particular to agree with you. Vaguely similar to saying "Twitter shows me a lot of people agreeing with me", the algorithm does that because people like it.

Here's an example that I think garners more agreement that properties of a limit ("really understanding") don't necessarily mean that any path towards that limit has the properties of the limit. I think there's a lot of room for disagreement about whether this is a factually-accurate analogy and I'm not trying to argue either way on that, just trying to answer your question about how one might make these sorts of arguments if one has a certain belief about the facts.

Let's say we're trying to build a calculator that only needs to do integer addition. And we decide to build it by building a giant if-else chain that hardcodes the answer to each and every possible addition. And due to finite resources, we're going to hardcode all the additions of integers up to absolute value N, but we will increase N over time.

Everything you said applies equally to this situation: it quacks like a duck, and when we talk about things it can't do we have to continually move the goalposts each time a new version comes out. It also has the property that there is a "scaling law" that says that each time you double N you get predictably better performance from the system, and you can do this without bound and continually approach a limit where it can answer any question indistinguishably from something we might call "true understanding".

But I think it's a bit easier to agree that in this case that it's not "really doing" addition and is a bit short of our wish to have an artificial addition system. And if someone touts this system as the way to automate addition we might feel a bit irritated.

Again, many people will say that this is a bad analogy because LLMs operate quite differently, and I'm not trying to argue for or against that. Just trying to give my explanation for how a certain understanding of the facts can imply the kind of conclusion that you are trying to understand.

I'm a bit rusty on the details, but I think the general reason that GPUs have less memory than you want is because they are typically optimized for throughput, which limits capacity in a couple ways. One is higher power consumption, and another is low max distance between processors and memory.

We already have the ability to put far more memory in GPUs than they currently have, but we don't because it hurts throughput. I don't think technical advances will change the existence of a throughput/capacity tradeoff, but I do think it's an interesting idea that maybe DL models benefit from a different tradeoff than graphics applications. Personally I'd guess that's a larger factor here than incremental changes in capabilities, because the two types of memory are both improving at the same time. But who knows! Maybe the technologies will converge enough that it will no longer make sense to have two different variants, though I wouldn't bet on it.

These articles/letters seem to just say "this is an important issue and I really hope someone else solves it!" and feel like passing the buck to me. Maybe it's more defensible from a journalist/author, but even that recent prominent letter didn't have any concrete suggestions and just called on other people to solve things.

I know there are people working on AI safety, but that's been going on for a while and it seems like the state of the art is to just think of things that we hope the AI doesn't do and then we tweak it to be unlikely to do it? It seems like we don't even know what to ask for when it comes to AI safety so people just ask for dialog/debate.

So I'm starting to feel cynical about our ability to actually achieve any of the safety that people wish for. This has been a known issue for a while, and despite a lot of awareness we haven't really made much progress on it.

I'm not sure what to do with this -- personally I feel a bit fatalistic that AI will continue unabated and we won't figure out the safety angle. But I just think the conversation should move on from "we really need to talk about this" because people certainly have been talking about it.

I think the silver lining of loneliness being so prevalent is that other people are feeling it too and want to do something about it. Knowing nothing about your situation I would guess that statistically if you're feeling lonely then your friends are also likely to feel lonely -- maybe they also say that their friends don't call them. I think this is a great opportunity in that there's desire from other people to connect with you that you could tap into.

Speaking from my own experience, it's very uncomfortable to be forward about trying to make friends / strengthen friendships. It reminds me of the stigma that online dating had when I first did it circa 2010 -- it has a connotation of being "desperate", and maybe that cool people wouldn't have to be desperate. But in my personal experience I've seen this shift a lot in the last year, where there is a lot of relief and appreciation when I mention wanting more friends and hint that I want to be friends with the person I'm talking to (they often respond by saying that they want more friends too). So I personally decided that I would rather risk looking desperate than feel lonely, and I recommend that tradeoff, especially because it's not perceived negatively like we worry it will be.

That's not to say that I think there's some easy option that I think you're avoiding; from my experience it has taken a number of incremental steps (and it's not like I've fully "solved" it). I guess I'm just trying to say don't get discouraged by the magnitude of the situation -- it might seem like a single phone call to a friend could never make a difference, but things like that add up over time.

As for practical suggestions, I've found video games to be a good activity to do with people that I don't live near. I've also found Buddhist meditation to be a great option for this -- it has the same sort of social structure and benefits as organized religion, but without requiring any particular sorts of beliefs or devotions. A lot of that has moved online these days -- I haven't tried this out but there are things like this https://www.sit-heads.com/

There was a post on here maybe a few months ago (couldn't find it right now) about how a certain percentage of people will just answer polls randomly/adversarially/jokingly, and I think the number they gave was something like 7% (I think they ran a poll with one nonsense answer and 7% of people picked it, something like that). Anyway that was just one data point, and I'm sure there are cross-cultural differences, but I think the idea still stands that there is an absolute margin of error in polling that is independent of the true percentage and sample size.

I also discovered a parallel effect that at least in my case is a bigger issue: if you strain while thinking, you can end up engaging head muscles which can fatigue quickly and become painful. For me, thinking hard hurts and I thought it was a brain thing, but now I think it's primarily strain-related tension headaches, which oddly can feel like a brain thing.

So from my experience there seems to be a possibility that when you think you are having brain tiredness you're actually having a physical headache.

I said this when FTX collapsed: I think FTX at least made a token effort to be legitimate, and Binance does not. Binance just hadn't gotten caught yet -- I think they will end up being worse than FTX. In behavior at least; the crypto ecosystem has a very high tolerance for bad behavior, so they might not see the crisis of confidence that FTX did.

I'm curious about the higher level issues at play as well, which I'm just guessing about since I only have a lay understanding of the field

Seems like there's probably a "winners curse" variant: conditioned on a bid being accepted, it is likely to be underestimated even if unconditioned there's an equal probability of under and over resonating

Also seems like there's an incentive alignment problem as well: there's a clear benefit to bidding lower, and if the costs of underestimating are not born by the bidder then you would expect to see systematic underbidding even if we have on-average-accurate ways of estimating.

These feel like important issues that are orthogonal to our ability to estimate things accurately and I wonder how much of the effect is from which type of cause (I wish the author had talked about this)

Before most of our time, but my understanding is that the past restrictions on cryptography research and export were reasonably effective?

Yes they were disliked and probably with good reason, but just mentioning it as a counterpoint that perhaps it is possible. You could have made a similar argument that anyone with pen and paper could do cryptography

Personally I think there's no way to be sure other countries aren't doing it, so perhaps it will continue in some government controlled way

Disclaimer: developer on Pyston, which could be considered a competitor

My concern is: there have been a few projects already that are, from the outside, more or less the same approach and set of tradeoffs as this. And they haven't been that successful. Given that this is treading familiar ground I would expect some words about how this is different, and the lack thereof makes me a bit skeptical to say this will become successful when others did not.

$100m future dollars, which are less valuable than present dollars.

Many types of institutions have to mark to market on a ~constant basis, so it's not impossible. Yes there are certain asset classes (private companies, for example) that don't have easy or reliable marks (but people still do it anyway!). But at least for SVB the issue was not determining the market value, but that the market value was bad.

Welcome to a non-zero interest rate environment. "$100m equivalent today" is not $100m -- the term to search for is "net present value". These considerations are precisely what marking to market captures

IANAL, but here's some FTC guidance on this, which I read (again IANAL) to say that the transactional-vs-marketing determination is not up to the sender but up to the recipient and if they "reasonably... would likely conclude" that it is marketing. Unfortunately the guidance is just about the subject line, so Experian might be toeing the letter of the guidance.

https://www.ftc.gov/business-guidance/resources/can-spam-act... (see the section "Q. What if the message combines commercial content and transactional or relationship content?")

I'm not quite sure what the legal enforceability is of the this sort of guidance, but at least with small senders when they try to pull this trick I send them this page and then I no longer receive any of these emails from them.

It sounds like you're saying "if you don't sell you can't lose money"

IANAL but I think yes and no? I think you're still obligated to pay your loan payments, but there's also a concept of "netting" where in certain circumstances you are entitled to what you describe.

Definitely don't take my word for it, just throwing this out there as maybe a starting point for more research.

I don't think anyone is saying this? The decrease in value is due to wrong-way interest rate exposure, not due to a fire sale

These statements seem pretty contradictory:

- "it had reserves in excess of its liabilities" - "They can't unwind that position and cover all possible demands"

I'm guessing that you're thinking of some sort of valuation of their assets that says something like "well they're really worth more than they're currently valued at", which is a common claim on this story but it's a pretty bold one?

I mostly think this is a good thing, but there are some fields where I feel like employers will have to take other precautions that are not as nice.

Most of us here are in tech where "ideas are cheap" is mostly true and commonly thought, but there are other fields (in particular, finance) where single ideas can easily be worth millions of dollars. (Imagine something like "the stock market goes down on Friday the 13th's".) How are employers supposed to let employees know these ideas? If there aren't noncompetes then I suspect employers will limit employees (and their growth) in other ways.

At least here in NY firms will pay ex-employees during their non-compete. My understanding though is that they only have to pay the base salary, which is usually a minority of the total compensation (even guaranteed compensation). Would be nice if that were fixed.

Having worked on this for a while, one way that might be helpful to understand this is that Python jits (such as the one I work on, Pyston) do in fact make Python code much faster, but the fraction of the time that is spent in "Python code" is only about 20% to begin with, with the other 80% being spent in the language runtime.

For example if you write `l.sort()` where l is a list, we can make it very fast to figure out that you are calling the list_sort() C function. Unfortunately that function is quite slow because every comparison uses a dynamic multiple-dispatch resolution mechanism in order to implement Python semantics.

Agree to disagree, I suppose. Another comment in this thread suggested using the value of pi to demonstrate intelligence, but you can imagine alien civilizations without the same reverence of pi as we do (if they are large enough). I mentioned in another comment that you can also imagine small enough aliens not inventing a real number line because they do not have the experience that everything is infinitely divisible.

I believe our difference is that you are presupposing that aliens will have essentially the same experience with reality that we do. I agree that if somehow a separate human civilization evolved in another galaxy, their math would likely look similar to ours. But you can imagine aliens that have very different experiences of "what actually works in the real world" (due to experiencing a different set of physics, such as at a different physical scale) and thus come up with different math.

Again, talking a bit outside my comfort zone, but I believe our universe's physics say that circles only have a fixed circumference-to-radius ratio at "small" (astronomically) scales. At large enough scales space is curved and the ratio begins to change.

In a similar way, a set of different physics would apply to small enough aliens that are principally governed by quantum mechanics. They might never come up with the real numbers because they do not believe that quantities are dense.

We could think of more examples -- aliens who typically move at speeds close to the speed of light would have very different conceptions of time, etc.

I really like the Stanford encyclopedia of philosophy. In this case: https://plato.stanford.edu/entries/philosophy-mathematics/

It's turtles all the way down -- we don't know what "truth" is, or how one can come to "know" truth (as opposed to just believe it). I'm less informed on these topics so I avoided them, but I've heard that our notion of truth is intimately tied to our experience of physical reality, and it seems easy to imagine aliens that have different experiences of reality.

For example, what if aliens are so large that the non-euclidian nature of space becomes obvious to them? Perhaps they would never come up with euclidian geometry. Maybe they would never create the notion of pi. Yes it shows up in other areas but you can certainly imagine an alien civilization that has a different circumference-to-radius ratio or disavows the entire idea of having a constant circumference-to-radius ratio.

As much as I like PG's writing in general, this is him dipping his toes in waters that are already well studied and coming off as ignorant and/or presumptuous.

His very first premise "The truths of mathematics would be the same, because they're true by definition" shows an ignorance of an entire field of discourse called the philosophy of mathematics. Whether aliens would have the same math as us is one of the fundamental questions, and to presume the answer shows a disdain of the existing body of thought on this topic.

In particular, we humans still hotly debate which math is the "right" math. Is the axiom of choice "true"? Does it make sense to say that it has a truth value? Is such a truth value amenable to human discovery? Does PG know the answers to all of these questions?

PG is either unaware of the human-choice nature of the fundaments of math or he thinks he has special knowledge of it. Either way makes this essay one of my least favorites.

> all this while the productivity they generate is at an all time high

I see this claim repeated a lot but you didn't say where it comes from so it's a bit hard to discuss. In particular, you are making a claim about the productivity of low earners but I don't think this is something that is measured in the US? My understanding is the BLS computes industry-wide productivity measures, ie mean productivity (as opposed to say median productivity), and one would expect this distribution to be significantly right-tailed and the mean to be mostly influenced by the right tail.

I haven't seen any data that breaks down productivity growth by income, so if you have a source for your claim I'd love to update myself. When I search for "productivity growth by income" I see versions of your claim but again they are all about population _means_.

Streamlit has completely replaced my usage of Jupyter -- I find it to have the quick iteration speed and visual output of notebooks, but it's just normal python so all the normal tooling works (there is no "git problem") and you don't have the weird state problems of notebooks.

Definitely recommend checking it out if you haven't already!