3 ms·

I’m not sure how to put this gently, but if you think computer chess is the same thing as LLMs, you’re not qualified to have an opinion on the future of technol

by xk_id 3y ago

I’m not sure how to put this gently, but if you think computer chess is the same thing as LLMs, you’re not qualified to have an opinion on the future of technology.

ben_w 3y ago
Chess was clearly just an analogy, and of a specific (old fashioned) AI doing what was widely believed (before we solved it) to be a uniquely human indicator of intelligence; it's not like anyone can accurately list the things that we will actually invent over the next 5-50 years. Worse: thanks to psychology, even if a time traveller came back to now with such a list (and proof of being a time traveler), their examples would not feel real to most until they could be experienced directly — I knew that GPT-3 could code simple web pages just over two years[0] before ChatGPT came out, but it still didn't hit me emotionally until we were all able to play with it ourselves. [0] August 8, 2020: https://youtu.be/_x9AwxfjxvE https://youtu.be/_x9AwxfjxvE
- xk_id 3y ago
  An analogy is only useful if it is meaningful. IBM's Deep Blue had a fundamentally different design than LLMs. "AI" is just a broad, abstract category that they both happen to share. The other day I heard someone argue that being skeptical about the inevitable "blockchain rennaissence" is like being skeptical about smart phones, back in the day (in this case, the category was just "technology", which is even more vague). Statements like this are not grounded in reality, they are just abuses of logic, useful to influencers in fabricating trends.
  - ben_w 3y ago
    > IBM's Deep Blue had a fundamentally different design than LLMs. Yes, and? I mean, you clearly get what you put in the next sentence: > "AI" is just a broad, abstract category that they both happen to share. But, there's also no requirement for future chat bots to be Transformer models like GPT. All we know for sure is that this blob of weights seems to be pretty good at a lot of different things that used to be things that only humans could do. Thus, I think as another demonstration of "we used to use this as a proxy for intelligence and now we can't" it's fine to use chess as an analogy because it demonstrates exactly that thing. No idea what comes next, nor when — it's still possible that "solving" general intelligence is beyond human capacity.
    - xk_id 3y ago
      I've read science fiction literature which had more substance than this kind of reasoning. When the argument approaches "on a long enough time scale there is no limit to progress, so anything is theoretically possible,", then it becomes just a platitude not worth stating.

by xk_id 3y ago

I’m not sure how to put this gently, but if you think computer chess is the same thing as LLMs, you’re not qualified to have an opinion on the future of technology.

ben_w 3y ago

Chess was clearly just an analogy, and of a specific (old fashioned) AI doing what was widely believed (before we solved it) to be a uniquely human indicator of intelligence; it's not like anyone can accurately list the things that we will actually invent over the next 5-50 years. Worse: thanks to psychology, even if a time traveller came back to now with such a list (and proof of being a time traveler), their examples would not feel real to most until they could be experienced directly — I knew that GPT-3 could code simple web pages just over two years[0] before ChatGPT came out, but it still didn't hit me emotionally until we were all able to play with it ourselves. [0] August 8, 2020: https://youtu.be/_x9AwxfjxvE https://youtu.be/_x9AwxfjxvE

xk_id 3y ago
An analogy is only useful if it is meaningful. IBM's Deep Blue had a fundamentally different design than LLMs. "AI" is just a broad, abstract category that they both happen to share. The other day I heard someone argue that being skeptical about the inevitable "blockchain rennaissence" is like being skeptical about smart phones, back in the day (in this case, the category was just "technology", which is even more vague). Statements like this are not grounded in reality, they are just abuses of logic, useful to influencers in fabricating trends.
- ben_w 3y ago
  > IBM's Deep Blue had a fundamentally different design than LLMs. Yes, and? I mean, you clearly get what you put in the next sentence: > "AI" is just a broad, abstract category that they both happen to share. But, there's also no requirement for future chat bots to be Transformer models like GPT. All we know for sure is that this blob of weights seems to be pretty good at a lot of different things that used to be things that only humans could do. Thus, I think as another demonstration of "we used to use this as a proxy for intelligence and now we can't" it's fine to use chess as an analogy because it demonstrates exactly that thing. No idea what comes next, nor when — it's still possible that "solving" general intelligence is beyond human capacity.
  - xk_id 3y ago
    I've read science fiction literature which had more substance than this kind of reasoning. When the argument approaches "on a long enough time scale there is no limit to progress, so anything is theoretically possible,", then it becomes just a platitude not worth stating.