HN

Comments (22)

• Menu

  ux266478
  That's a mischaracterization. Latent space is simply a (multidimensionally) sorted collection, it's only a piece of the pie. A massive amount of structure is held in the unembedding layer. Generative AI models are a very specific ordering, LLMs a very specific subset of that, and they're hardly the only users of the concept.I get what the author is going for, and they're on the right track. There is something interesting going on with embedding spaces: When used as the substrate for a neural network, you can effectively treat them as a kind of continuous form of computation. That is, given two functions, you can trivially derive a function which sits exactly between those two, and do so ad infinitum, for any arbitrary program (in theory. Obviously everything materially accessible is finite.) This is only one such manipulation. You can deform a function in an unenumerable amount of ways. Think like a bezier curve path tool in something like Krita or Photoshop, but for a function. You can keep adding points and twist it to your heart's content.It's wrong to focus on LLMs specifically, as well. This is a much, much broader topic than you realize. Most of the interesting stuff has nothing to do with language models at all. I get a huge chunk of the industry is currently having a stroke over LLMs being able to brute-force problem solving, but if we're to talk philosophy, theory, and so on, we have to get past the surface level misuse of Machine Translation's holy grail. That's like having a conversation about the potential of computation itself, but all you talk about is web browsers, using them interchangeably with "computer".
• chroma_zone
  There's a lot of very strong claims made at the start of this article.(emphasis mine)> A Large Language Model (LLM) is like a small zip file that contains all human knowledge.> In a strange but real way the resulting tiny file contains all the information that is on the internet and in our libraries.> Likewise, the LLM could recognize the face of almost any person, and it could generate any possible human face,All writings? All of human knowledge in general? Any person??The example he gives for writing is Shakespeare, which might be one of the most overrepresented writers in the entire training dataset. So yes, of course LLMs can replicate his writings with high accuracy. That doesn't mean that the same applies to literally all of human writings and knowledge.> We've never had a system to integrate everything we know and everything we can imagine.Yeah, we still don't.At first I thought he was just being intentionally hyperbolic for effect, but the rest of the article is even worse. From the closing paragraph:> We’ll soon depend on this oracle to such an extent that we’ll wonder how we lived without it.No! AI is not an oracle! That is honestly an extremely dangerous way to think about this technology.This is the type of baseless hype that OpenAI and Anthropic have been exploiting for years, and I really wish it would stop.
• Menu

  luisln
  >In other words, correctness, truth, cohesiveness, completeness, comprehension, etc are all essentially patterns that are mapped in this space.The caveat is that truth in latent space is just a reflection of the consensus from the corpus, and you find truth by comparing the answer in latent space to what is in reality.But I just hate this idea that truth and facts are no longer real, they're just "directions". The more we rely on these models for our lives, the more we lose touch with reality and are pushed and pulled in all these different directions. Feels like the future is just ai psychosis and there's no way out. Is that what complete agi victory looks like?
• Menu

  invictati
  Dear Kevin: AI models don't contain all of human knowledge. They don't even contain, for example, the complete curricula of the least comprehensive K-12 program.
• jschveibinz
  Probably the most important thing I've read this month. The key concepts are valuable.
• hmokiguess
  This post made me want to watch Everything Everywhere All at Once again
• umm4gemm4
  Some math related to training on latents, not tokens:https://arxiv.org/abs/2605.27734
• worldthruword
  In future, to provide guard rails to models, we will build token flow pipes in mathematical sense. Sort of like Mathematical Engineering.
• Menu

  inigyou
  Infinite Craft is a web game based on combining two things using LLMs, but it's not clear how it works, whether latent space or something else.
• anon
  undefined
• yapyap
  what an awful, nonsensical article
• dannyw
  This is nonsensical AI slop with so many technical mistakes.
• lardosaurusrex
  This article is written like someone sat around and wrote out as many potential buzzwords as possible and then came up with definitions after the fact.