6 ms·
This was a reply to another comment on how uncanny it is that gaming hardware allowed the current machine learning advancements. It was deleted before I finishe
by MrLeap 3y ago
This was a reply to another comment on how uncanny it is that gaming hardware allowed the current machine learning advancements. It was deleted before I finished writing, but I'm not letting that stop me from sending this anyhow.
Creative immersive experiences, across the history of computing, are computationally expensive. The better computing gets, the more (visually at least) immersive the experience. The search for photo realism is an asymptotic search to compute a convincing simulacrum of reality. So much human nature drove the tip of that spear forward.
Immersion itself implies some kind of tricking of the consciousness, blending the familiar with the novel. Instructions and ideas piped into a machine that generates images. GPUs are specialized in parallel computation because that's precisely the architecture that best performs the multi-modal magic trick.
It's surprising to me how incredulous people seem about the possibility of creating a sentient machine that is fed off a large enough diet of human written material. Language is crystalized consciousness. It's the primary medium we use to communicate our inner experience to the outside. What other fuel would you use?
- Scoundreller 3y agoI wonder what would have happened if 3dfx had survived. It ultimately had its assets acquired by nvidia! https://en.wikipedia.org/wiki/3dfx https://en.wikipedia.org/wiki/3dfx
- galeos 3y agoI've often wondered this. Would be interesting to see how useful that 3dfx IP ended up being? When I was a teenager I tried to get my parents to open a brokerage account so I could buy some 3dfx stock as I was so excited by what they were doing.
- fastball 3y agoThey failed because they were out-competed. Not really sure how the timeline would be much different if they'd managed to hang on a little longer.
- growingkittens 3y agoMy language skills are damaged by a childhood brain injury. I think in pictures; I remember information with emotions, music, movement, and basically anywhere my brain could stuff information. Words are ephemeral, I often forget the beginning of a sentence by the time I reach the end. The human brain is so much more than language.
- ChatGTP 3y agoThe sound of the rain requires no translation.
- Shorel 3y agoI completely agree with you. The fact that we can take ideas and serialize them into language doesn't mean the language and the ideas are the same thing. They become ideas again when we deserialize them into our minds. Language models can't do that (something else more complex could, but then it would not be just a language model).
- growingkittens 3y agoThe next generation of AI must be more than a language model. I believe AI will reach a stagnation point otherwise. Also slowing down AI is that: 1) We don't presently know that much about the human brain or psychology. A lot, yes, but relatively not that much. 2) The scientific method is systematic, not systemic. Both perspectives are needed to interpret a reality made of interconnected systems. 3) There are many specialists and not many generalists. Generalists are needed to tie together the multiple domains required to emulate the human brain.
- schultzie 3y agoCan I message you about your brain injury experience? Your description of what your life/experience matches my own, and I had several concussions as a child. I never considered the possibility my "unique" way of remembering things and speech patterns was linked to brain trauma.
- 3y ago
- PoignardAzur 3y agoThat doesn't really track. There's no obvious link between video games and machine learning, it's just that video games graphics were the one sector demanding low latency sky-high performance at the time the hardware became possible, so that's where the money went. If video games didn't exist and instead we had found some extremely valuable ore that could only be mined by performing lots of hash calculations (so like bitcoin, but actually valuable), machine learning would have developed on miner asics, and then people would be telling stories about how of course sentient machines could only come from hashing numbers together because hashing is what our brain does or whatever.
- seanthemon 3y agoFrankly, the link is linear algebra and matmul operations, exactly what's used to drive cameras, movement and much more in video games.
- tnecniv 3y agoAs someone else said, not really. The link is linear algebra which is perhaps the best computational tool we have. In many areas of applied math, as soon as things get nonlinear (or non-quadratic if we’re taking derivatives since those derivatives become linear), it is impossible to find closed-form solutions (I.e., solutions you can write in a simple form). A big reason for this is that solving Ax = b might require a matrix (pseudo)inverse, but once you have that inverse, you have the solution for any pair (x, b), whereas if you had to solve f(x) = b, a new pair requires a fresh computation. Put in other terms, for linear problems, local properties are often global properties when studying linear problems. A lot of the numerical methods we use to solve hard problems are based on solutions to linear problems (or quadratic problems if we’re taking derivatives) that predate computer graphics because those problems are simple to solve. We can just solve them a lot faster now so approximating nonlinear things with linear things iteratively becomes a lot more computationally tractable.
- Shorel 3y ago> Language is crystalized consciousness. That's a pretty big claim. I disagree with it, but I agree about the idea of language models being a fundamental part of an artificial mind. Language is like the TCP/IP of ideas, we put data into it, and then extract data from it. But the language and the data we extract from it are two different things, IMO.