Elektrine lite

← Feed

Nathan Harvey

NathanHarvey@kolektiva.social

<p>Father. Aspiring ally, publishing models on structuralized bigotry and power. Lead of the graphics kernel team at Microsoft.</p>

Posts

  • View post

    @emilymbender Dr. Bender, I fear you have been spreading a lot of misinformation about these models and what they do, and your views are showing their errors here. LLMs don’t model languages. They aren’t string manipulators in the sense you imply. LLMs model the “language producing processes” they are trained on. Training is semantic. Every researcher making claims about the meaning produced by LLMs should be very clear about why these models are able to produce output with such semantic stabi...

  • View post

    @walruslifestyle@octodon.social @dpiponi @gregeganSF @BoydStephenSmithJr Although I stopped following Wolfram long ago due to issues I saw in his mathematical and popular work, I will accept he may have described AI like you state. It is wrong, though. What you describe is an n-point autocomplete. That process is not a trained model process, but a simple statistical analysis on large text collections. It only loses information by summarizing statistical moments of data and used as a statistical...

  • View post

    If you live in the Vancouver, British Columbia area and are looking to play a part in the great advancements occurring in GPUs, NPUs, and other accelerator devices, our team has a great senior-level opportunity! Our team works in the Windows kernel to develop the driver abstraction model and designs the submission scheduling architecture for the device engines, manages the device memory, virtualizes the device into VMs, manages the power policies, and generally provides the glue between the runt...

  • View post

    @foolishowl @StefanThinks The latest one I saw was from Monday, but it happens so often I suspect it’s not the latest…

  • View post

    @julesh @zanzi Tarski also famously spent some time in crucial papers on early model theory discussing the finite symbols issue.

  • View post

    @BartoszMilewski @johncarlosbaez @dpiponi This repeatability claim is not quite right. What is true is that statistical tests of data conformance are model-dependent. You can check predictive models against new data using common test protocols, and this will tell you the significance of that data in bisimulating the model. In models with time translation symmetry, this translates to “repeatability” requirements on the test. But ultimately, models of finite universes do not have time translatio...

  • View post

    @yosh Elizabeth Warren worked with Emmanuel Saez and Gabriel Zucman to build her wealth tax proposals. These two economic researchers have been vocal proponents for the tax as a resolution to the problematic dynamics of wealth accumulation through economies of scale. They have a famous paper “Progressive Wealth Taxation” which goes over much of the reasoning and data for a wealth tax, and they have a more popular-oriented book “The Triumph of Injustice” which is a higher level look at tax policy...

  • View post

    @VCP @funnymonkey @GossiTheDog Those are the chains of Jacob Marley. He had them gold plated.

  • View post

    @highergeometer @johndcook I think the best explanations of this take you on the semantic journey: you have a syntax around equality and are looking for a model to interpret it. This is a fascinating task in the abstract, and there is a lot of great math and philosophy just around concretisation and distinctions of isomorphism and equality, and it gets pretty wild once real world considerations of distinguishability are introduced. When looking to get students fascinated by interesting equalit...

  • View post

    @johncarlosbaez John, I never liked when you used to mock intuitionists on usenet. I had thought that over the years since, as you had realized the deep mathematical underpinnings of the philosophy and its relationship with semantic views of math, you had grown more open to philosophical debate more generally in foundations. Personally, as I’ve followed Wildberger over the years on YouTube and elsewhere, I’ve noticed he spends considerable effort detailing his philosophy and its reasons. He’s a...

  • View post

    @highergeometer @TobyBartels @sibrosan When many radical constructivists use the word “meaningful”, they mean this formally as having reference (in reality or experience). Some numbers have natural interpretation in real experience - counting and fractions are processes we participate with in the world, which is what makes them “meaningful”. But no one ever actually does infinite processes. No one has ever experienced them. This idea of “meaningfulness” is critical to the syntax/model distincti...

  • View post

    @TobyBartels @highergeometer @sibrosan It can be hard to talk of metamathematics in ways that don’t reify unobserved absolutes, but let me try from the opposite direction. In folk language, we know what we mean when we ask for half of something. There is a vague notion of transforming that thing into something countable and then splitting the counts in some manual way. The semantics have those beautiful fuzzy boundaries of folk languages, but we talk about halves and quarters all the time and...

  • View post

    @highergeometer @TobyBartels @sibrosan I wouldn’t be able to accurately represent Wildberger’s objections to sqrt(2), but I can start with the fifth root of 7 to give a better idea here. Well, first actually let me start again with fractions. With fractions, we have finite processes to give us real world implementations of them. For interpretations on distance, for example, we have the classic straightedge and compass techniques. We can show that a finite set of steps is expected to give an int...

  • View post

    @highergeometer @TobyBartels @sibrosan I think one of the reasons I don’t like the mockery is that it ultimately tends to build environments where it is easy to attack people for things they don’t believe. It’s absurdifying. And I don’t think you’re intentionally lying about Wildberger, but I think you are able to uncritically believe these absurdifications, that you haven’t verified them yourself but they fit the environment of mockery and so make sense to you. I think anyone who actually has...

  • View post

    @johncarlosbaez @j_bertolotti You are on a mockery tear lately. For people interested in what “experts” think and eager to exclude the authors for some unstated reasons (likely tied to the environment of mockery), physicists might be interested in reading other works unstated by Baez that show similar results, like https://arxiv.org/pdf/2402.13166 Certainly, though, bloodlust and mockery are the zeitgeist of the world, and just agreeing with derision can often be the easiest course.

  • View post

    @matt This image has two graphs. On the top of the image is a graph of the relative absorption coefficient of water. This has an x-axis representing the wavelength of the incident light in meters, as a logarithmic scale from 1 to 10^(-12). The y-axis is the absorption coefficient, again as a logarithmic scale, from 10^(-3) to 10^6. The graph starts low at under 10^(-3) for 1 m light, but quickly rises to around 10^1 at wavelengths around 0.1 meters. This is marked “Rotation” on the graph. The...

  • View post

    @keenancrane Although it is common to think of autoencoders “seeking to reproduce their input as closely as possible”, this isn’t actually their design. The scoring function is not always (maybe not even typically these days) a distance / metric over the input space. We know it is impossible to generally reproduce exactly (the information content in this process is lossy), so we don’t require exactness, but also the information content typically has other important qualities. So modern autoenc...