20 comments

  • helloplanets12 hours ago
    For the visual learners, here&#x27;s a classic intro to how LLMs work: <a href="https:&#x2F;&#x2F;bbycroft.net&#x2F;llm" rel="nofollow">https:&#x2F;&#x2F;bbycroft.net&#x2F;llm</a>
  • tpdly12 hours ago
    Lovely visualization. I like the very concrete depiction of middle layers &quot;recognizing features&quot;, that make the whole machine feel more plausible. I&#x27;m also a fan of visualizing things, but I think its important to appreciate that some things (like 10,000 dimension vector as the input, or even a 100 dimension vector as an output) can&#x27;t be concretely visualized, and you have to develop intuitions in more roundabout ways.<p>I hope make more of these, I&#x27;d love to see a transformer presented more clearly.
  • vivzkestrel2 hours ago
    - while impressive, it still doesnt tell me why a neural network is architected the way it is and that my bois is where this guy comes in <a href="https:&#x2F;&#x2F;threads.championswimmer.in&#x2F;p&#x2F;why-are-neural-networks-architected" rel="nofollow">https:&#x2F;&#x2F;threads.championswimmer.in&#x2F;p&#x2F;why-are-neural-networks...</a><p>- make a visualization of the article above and it would be the biggest aha moment in tech
  • esafak13 hours ago
    This is just scratching the surface -- where neural networks were thirty years ago: <a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;MNIST_database" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;MNIST_database</a><p>If you want to understand neural networks, keep going.
    • abrookewood4 hours ago
      Which, if you are trying to learn the basics, is actually a great place to start ...
  • brudgers2 days ago
    The original Show HN, <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=44633725">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=44633725</a>
  • swframe26 hours ago
    This Welch Labs video is very helpful: <a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=qx7hirqgfuU" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=qx7hirqgfuU</a>
  • chan14 hours ago
    Super cool visualization Found this vid by 3Blue1Brown super helpful for visualizing transformers as well. <a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=wjZofJX0v4M&amp;t=1198s" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=wjZofJX0v4M&amp;t=1198s</a>
  • ge9611 hours ago
    I like the style of the site it has a &quot;vintage&quot; look<p>Don&#x27;t think it&#x27;s moire effect but yeah looking at the pattern
    • Bengalilol8 hours ago
      Lucky you!<p>&lt;<a href="https:&#x2F;&#x2F;visualrambling.space&#x2F;dithering-part-1&#x2F;" rel="nofollow">https:&#x2F;&#x2F;visualrambling.space&#x2F;dithering-part-1&#x2F;</a>&gt;<p>&lt;<a href="https:&#x2F;&#x2F;visualrambling.space&#x2F;dithering-part-2&#x2F;" rel="nofollow">https:&#x2F;&#x2F;visualrambling.space&#x2F;dithering-part-2&#x2F;</a>&gt;
      • ge968 hours ago
        Oh god my eyes! As it zooms in (ha)<p>That&#x27;s cool, rendering shades in the old days<p>Man those graphics are so good damn
  • 8cvor6j844qw_d69 hours ago
    Oh wow, this looks like a 3d render of a perceptron when I started reading about neural networks. I guess essentially neural networks are built based on that idea? Inputs &gt; weight function to to adjust the final output to desired values?
    • mr_toad4 hours ago
      The layers themselves are basically perceptrons, not really any different to a generalized linear model.<p>The ‘secret sauce’ in a deep network is the hidden layer with a non-linear activation function. Without that you could simplify all the layers to a linear model.
    • sva_5 hours ago
      A neural network is basically a multilayer perceptron<p><a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Multilayer_perceptron" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Multilayer_perceptron</a>
    • adammarples7 hours ago
      Yes, vanilla neural networks are just lots of perceptrons
  • jazzpush29 hours ago
    I love this visual article as well:<p><a href="https:&#x2F;&#x2F;mlu-explain.github.io&#x2F;neural-networks&#x2F;" rel="nofollow">https:&#x2F;&#x2F;mlu-explain.github.io&#x2F;neural-networks&#x2F;</a>
  • jetfire_17118 hours ago
    Spent 10 minutes on the site and I think this is where I&#x27;ll start my day from next week! I just love visual based learning.
  • 4fterd4rk13 hours ago
    Great explanation, but the last question is quite simple. You determine the weights via brute force. Simply running a large amount of data where you have the input as well as the correct output (handwriting to text in this case).
    • ggambetta13 hours ago
      &quot;Brute force&quot; would be trying random weights and keeping the best performing model. Backpropagation is compute-intensive but I wouldn&#x27;t call it &quot;brute force&quot;.
      • Ygg212 hours ago
        &quot;Brute force&quot; here is about the amount of data you&#x27;re ingesting. It&#x27;s no Alpha Zero, that will learn from scratch.
        • jazzpush29 hours ago
          What? Either option requires sufficient data. Brute force implies iterating over all combinations until you find the best weights. Back-prop is an optimization technique.
          • Ygg251 minutes ago
            In context of grandparents post.<p><pre><code> &gt; You determine the weights via brute force. Simply running a large amount of data where you have the input as well as the correct output </code></pre> Brute force just means guessing all possible combinations. A dataset containing most human knowledge is about as brute force as you can get.<p>I&#x27;m fairly sure that Alpha Zero data is generated by Alpha Zero. But it&#x27;s not an LLM.
  • cwt13711 hours ago
    This visualizations reminds me of the 3blue1brown videos.
    • giancarlostoro11 hours ago
      I was thinking the same thing. Its at least the same description.
  • atultw2 hours ago
    Nice work
  • shrekmas7 hours ago
    As someone who does not use Twitter, I suggest adding RSS to your site.
  • artemonster10 hours ago
    I get 3fps on my chrome, most likely due to disabled HW acceleration
  • anon29110 hours ago
    Nice visuals, but misses the mark. Neural networks transform vector spaces, and collect points into bins. This visualization shows the structure of the computation. This is akin to displaying a Matrix vector multiplication in Wx + b notation, except W,x,and b have more exciting displays.<p>It completely misses the mark on what it means to &#x27;weight&#x27; (linearly transform), bias (affine transform) and then non-linearly transform (i.e, &#x27;collect&#x27;) points into bins
    • titzer10 hours ago
      &gt; but misses the mark<p>It doesn&#x27;t match the pictures in your head, but it nevertheless does present a mental representation the author (and presumably some readers) find useful.<p>Instead of nitpicking, perhaps pointing to a <i>better</i> visualization (like maybe this video: <a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=ChfEO8l-fas" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=ChfEO8l-fas</a>) could help others learn. Otherwise it&#x27;s just frustrating to read comments like this.
  • pks01610 hours ago
    Great visualization!
  • javaskrrt11 hours ago
    very cool stuff