• 0 Posts
  • 82 Comments
Joined 7 months ago
cake
Cake day: July 10th, 2024

help-circle



  • I said “inspired by” and not “exact digital replicas”.

    In classical MLP networks a neuron is modeled as an activation function depending on its inputs. Connections between those are “learned”, basically weights which determine the influence of one neuron’s output on the next neuron’s input. This is indeed Inspired by biological neural networks.

    Interestingly, in some computer vision deep learning architectures, we have found structures after the training procedure which are even similar to how human vision works.

    There are a bunch of different artificial neural network types, most – if not all – inspired by biology. I wouldn’t be so bold to reduce them in that absurd manner you did.





  • The position with the vegan cats is basically indefensible.

    What do all organisms, including animals, need to properly maintain their metabolism?
    Nutrients.
    What are nutrients?
    A bunch of different chemicals.

    Depending on the specific organism, another set of nutrients is required, also varying in amount of course.

    All required nutrients for humans at least can be obtained or synthesized from non-animal compounds.

    From that simplified perspective, it’s absolutely rational to explore how we could feed animals like cats on a purely vegan diet.
    But it’s certainly nothing which should be left to do for the layman alone, as veterinarian care is advisable if harming the animal should be avoided.






  • If we’re speaking of transformer models like ChatGPT, BERT or whatever: They don’t have memory at all.

    The closest thing that resembles memory is the accepted length of the input sequence combined with the attention mechanism. (If left unmodified though, this will lead to a quadratic increase in computation time the longer that sequence becomes.) And since the attention weights are a learned property, it is in practise probable that earlier tokens of the input sequence get basically ignored the further they lie “in the past”, as they usually do not contribute much to the current context.

    “In the past”: Transformers technically “see” the whole input sequence at once. But they are equipped with positional encoding which incorporates spatial and/or temporal ordering into the input sequence (e.g., position of words in a sentence). That way they can model sequential relationships as those found in natural language (sentences), videos, movement trajectories and other kinds of contextually coherent sequences.