@glyph Did you quote post something?
-
@glyph There are even more layers!
My wife is a union Carpenter. Her main work for the last few years has been on teams of highly skilled construction workers building and rebuilding clean rooms to house the machines that are involved in chip making. Now she has a job building a new facility for a company that makes the machines - a project that takes hundreds of people (not to make the machines, just to construct the new facility).
Yup, it's turtles all the way down.
-
-
-
-
-
@glyph Worth noting - these machines also depend on some rather recent physics to even be plausible. So, there's also a layer of physics, material science, etc. academics at mostly universities, scattered across the world, mostly funded by public spending on sciences, to even get to the point the engineers could try to build something that works at all on this scale.
-
@glyph Worth noting - these machines also depend on some rather recent physics to even be plausible. So, there's also a layer of physics, material science, etc. academics at mostly universities, scattered across the world, mostly funded by public spending on sciences, to even get to the point the engineers could try to build something that works at all on this scale.
@miss_rodent yeah the video is partially a story about that part of the process, or at least a window into it. apropos of the point in my linked thread, even if LLMs could do this kind of reasoning (they can't) we are a thousand years and dozens of general breakthroughs away from a context window large enough to even hold the required prerequisites to start reasoning out these problems even in just the AI tech itself
-
@miss_rodent yeah the video is partially a story about that part of the process, or at least a window into it. apropos of the point in my linked thread, even if LLMs could do this kind of reasoning (they can't) we are a thousand years and dozens of general breakthroughs away from a context window large enough to even hold the required prerequisites to start reasoning out these problems even in just the AI tech itself
@glyph Oh, yeah, LLMs are ... idk, this hype wave jsut makes me think of ELIZA (1966 simulated therapist), really.
With about the same chance of achieving sapience. -
@glyph Oh, yeah, LLMs are ... idk, this hype wave jsut makes me think of ELIZA (1966 simulated therapist), really.
With about the same chance of achieving sapience.@glyph Complete with ELIZA effect....
https://en.wikipedia.org/wiki/ELIZA_effectThis just feels like one of the dumbest 'those who don't learn from history are doomed to repeat it' situations I've been alive for.
-
@glyph yeaaaah
-
-
-
@glyph @mcc then it becomes potentially easier. You would have to source some of the chemicals, which may not be produced in NA anymore but eh. That is probably manageable.
Note that if you Manhattan project it, you may be able to produce a few batch but you will lack the repeatable good yield. That needs experience that need time.
-
@glyph DUDE Thank you for linking this: https://theoutpost.ai/news-story/major-study-reveals-ai-benchmarks-may-be-misleading-casting-doubt-on-reported-capabilities-21513/
I've been angry about yet another "look, the benchmark says this LLM can do {x} amazing thing!" (with spurrious claims and lots of caveats) for a full day.
-
@glyph I really like your point about scaling down. Even when I was in undergrad, there were already pieces about how the long free lunch was finally over, but damned if the strategy of relying on hardware progress for all software improvements isn't entirely unsustainable.
-
@glyph great read, always a delight when people make well-argued claims for not trying to go for unreasonable scales!
-
-
@glyph I really like your point about scaling down. Even when I was in undergrad, there were already pieces about how the long free lunch was finally over, but damned if the strategy of relying on hardware progress for all software improvements isn't entirely unsustainable.
-
-
@ireneista @cthos @xgranade this isn't exactly a "cheerful" thought, but it's also not horribly grim: I think we already saw it break down in 2020, and we saw both how brittle it is (nobody had enough slack in their supply chain to actually weather the disruption without exposing catastrophic delays to customers) but also its resilience (customers were super mad, alternate pathways DID come online in less than a year)