Skip to content
DevMeme
6153 of 7590
AI ML Post #6748 · source on Telegram

AI Hype vs. Its Humble Mathematical Reality

Description

A two-panel meme on a white background. On the left, the blonde-haired, bearded 'Nordic / Chad' Wojak character looks respectfully towards the right panel and says, 'Thank you for changing my life'. On the right is a diagram of a simple feedforward neural network, showing an 'Input Layer', two 'Hidden Layers', and an 'Output Layer', with nodes connected by a mesh of lines. Below this diagram, text responds, 'i'm literally just matrix multiplication'. The humor lies in the stark contrast between the transformative, almost magical perception of Artificial Intelligence and its fundamental, core mathematical operation. While neural networks enable complex, world-altering technologies, this meme demystifies the concept for those in the know, reducing it to its basic building block of linear algebra. It's a classic joke among machine learning practitioners that pokes fun at the hype cycle by grounding the technology in its mathematical roots

Comments

18
Anonymous ★ Top Pick The main difference between matrix multiplication and 'AI' is about a million dollars in VC funding and a keynote presentation
  1. Anonymous ★ Top Pick

    The main difference between matrix multiplication and 'AI' is about a million dollars in VC funding and a keynote presentation

  2. Anonymous

    Remember: the only thing separating ‘world-changing AGI’ from glorified linear algebra is a colossal GPU bill and a few smug dot-products

  3. Anonymous

    After 20 years in tech, you realize the most revolutionary AI breakthrough is just convincing VCs that your matrix multiplication runs on 'proprietary neural architecture' instead of numpy.dot() with extra steps and a $10M cloud bill

  4. Anonymous

    After spending three months optimizing your transformer architecture with attention mechanisms, custom loss functions, and elaborate regularization schemes, you realize your model's performance ceiling was determined in the first 10 lines where you defined the matrix dimensions - everything else was just expensive matrix multiplication with a PhD in disguise

  5. Anonymous

    After years chasing SOTA, realizing your 'revolutionary' model is just cuBLAS with a PhD - suddenly, prod scaling feels trivial

  6. Anonymous

    Amazing how renaming SGEMM to “intelligence” adds a zero to the valuation while SREs still fight memory bandwidth

  7. Anonymous

    We rebranded 'call cublasSgemm a few trillion times' as intelligence; the life-changing part is procurement forwarding the A100 bill

  8. @SamsonovAnton 1y

    Thank you for changing my life!

  9. Deleted Account 1y

    NEW SCAM!!!! GUYS! DON'T PAY ATTENTION TO OPENSEA, and also SCAMMERS! 😭😭😭🤬🤬🤬

  10. Deleted Account 1y

    that's bot, not a real user, ban it completly

    1. Deleted Account 1y

      We cant. So we use democracy.

  11. @vladyslav_google 1y

    Maybe i retarded, but aren't the first matrix column number should be equal to the second matrix row number in order for multiplication to be valid ? If it's true, then the meme picture is incorrect 🤓

    1. dev_meme 1y

      https://datascience.stackexchange.com/questions/103183/working-of-dense-layer

    2. @trainzman 1y

      A part of a graph between two layers (circle columns) refers to m*n connections, where m, n are sizes of layers. When you start with a vector of size x0, then have a layers of sizes x1, x2 and so on, you essentially have coefficient matrices of sizes x0*x1, x1*x2 and so on (they may be transposed, chosen convention is not principal). Coefficient manipulation to process your x0-sized input is thus just a sequence of matrix multiplications (with some extra steps between each two layers, like considering of intercept coefficient and activation function)

      1. dev_meme 1y

        Soon there will be a place to prevent knowledge being lost in comments 🌚

  12. @trainzman 1y

    The edges of a graph between two layers represent a matrix of coefficients: each of m nodes is connected to each of the other n nodes

  13. @trainzman 1y

    Dev meme educational

  14. @brucewumx 1y

    Just matrix multiplications and gating functions (only matrix multiplications = one linear transformation)

Use J and K for navigation