Thread

1. stratos123 · 1791107241 · focus · HN ↗

   LeCun also said back in 2022 that "if you train a machine, as powerful as it could be, your 'GPT-5000', on text", it will never be able to learn basic common-sense physics like that objects placed on tables will move along with them.

   1. LoganDark · 1791112315 · focus · HN ↗

      I keep wanting to use LLMs for creative writing that heavily involves physics like this, and it's been a definite struggle to say the least. I recently discovered that Gemini 3.1 Pro is the first model I've found to clearly beat the original November 2022 ChatGPT release in terms of implied physics. Man did the world really take its sweet time to get back here. I think it will continue to be a struggle until another genuine architectural shift happens -- it's still not anywhere close to perfect, just better.

      1. wartywhoa23 · 1791113407 · focus · HN ↗

         [dead]

         1. LoganDark · 1791132028 · focus · HN ↗

            LSD is great!

            Jokes aside, no I'm not saying anything about creativity and LLM coexisting in one sentence. I genuinely try to use them for writing and I genuinely run into issues with other models missing details, and misunderstanding poses, or anatomy, or directionality, etc. I'm not hating on them for anything related to the term LLM (or creativity) but rather for the real issues that I've seen myself using them personally.

            So I'm saying Gemini 3.1 Pro is the best I've seen because it seems to be a decent bit better than frontier models at this. Genuinely. It seems better able to transfer concepts into less traditional areas, which is important when say, you have entirely non-human characters? (Which I always do.) A lot of models get stupid incredibly quickly in that case because they were trained with humans.

            1. wartywhoa23 · 1791133356 · focus · HN ↗

               > LSD is great!

               It is.

               It showed me that all human creativity and art is an attempt to express the indescribable Otherness in words, which always fails, and even the word "describe" in Russian literally translates as "write around" ("о-писывать"), and its close relative "define" means limiting, assigning an end to something infinite, thus leaving the essence outside of words.

               Now, LLMs operate totaly within words and hence will always be a parody of art.

               1. hardbass · 1791134437 · focus · HN ↗

                  So is that what it is? Words are another muddle btw unless you make it clear you only mean natural language and not any alphabet in general. Turing computers are also languages and as far as we know, can express anything in the universe. If you think you get some superpowers from drugs that enable you to access something outside your sense organs and brains input, that non drug users don't, show it. Eg if you think you can read what is happening in another room without any signal or leakage from there, you are free to demonstrate it. As far as we know drugs are not magic.

                  1. wartywhoa23 · 1791138663 · focus · HN ↗

                     > show it

                     I explicitly tell you that there are things (in fact, it's a single thing, fractally generating everything else) that can't be shown, expressed, or otherwise be reduced into language, and you keep demanding to show it, while in fact staring at it your whole life and failing to see.

                     Psychedelics (not "drugs" as you try to smear them) are just one way of the many to see it, but in the modern way of living, also almost the only one available.

                     1. hardbass · 1791140460 · focus · HN ↗

                        I am not smearing drugs. Your difficulty is proving somehow this beyond language thing exists. A drug affects your brain, lsd inhibits certain negative feedbacks in the brain, positive feedback tends to cause chaos which usually manifests as fractals. As of yet our brains are known to not follow any special laws beyond known physics, which is turing computable. If you think there is something "beyond" you have to prove it to an external observer.

                        1. LoganDark · 1791140655 · focus · HN ↗

                           Never seen this description of LSD inhibiting negative feedback in the brain. I've mostly seen reports that it increases connectivity in the white matter region. Have any resources about it?

                           Anyway, I think what they're saying is that if you train a model purely on generating language, it'll lack many of the things about human brains that result in the language they generate. The process can matter more than the result for language (specifically for creativity and art, too), so it follows that a model trained purely on the output is going to be missing something more fundamental, even when it does produce coherent language. This matches up with my LLM experience so far.

                           1. hardbass · 1791141211 · focus · HN ↗

                              I am not a mathematician or neurologist so do not take my word as granted but look into the work of LSD's interactions with the thalamus and 5-HT2A receptors. It is my understanding that LSD turns off an otherwise inhibited system so the feedback loop of certain neural signals goes from decaying to exciting leading to a highly sensitive state. But you should go ask a neurologist for better understanding.

Unofficial Hacker News client; not affiliated with Y Combinator.