3 ms·

Boy, people really love to ride the "AI is the new crypto" narrative, huh? Even pulling in spurious relationships to make the point (. Yet here I am, using this

Boy, people really love to ride the "AI is the new crypto" narrative, huh? Even pulling in spurious relationships to make the point (. Yet here I am, using this tech every day to for productive uses, greatly improving what I can accomplish. What did crypto ever do for us in comparison?

chx 3y ago
I am not participating in industrial strength gaslighting. "greatly improving what I can accomplish" is a trap. But to each their own.
- danielbln 3y ago
  You may be gaslighting yourself if you can't see the actual, tractable, current practical applications of this tech (and that includes local models you can run yourself).
  - chx 3y ago
    That's the thing: there is no actual and tractable use of this tech. It's probabilistic so every answer it can give can only resemble the real one and only accidentally be the real one so you need to fact check it anyways and then why bother? Do consider the AI written mushroom hunting guides being sold on Amazon. Would you trust these literally with your life?
    - flappyeagle 3y ago
      No but I don’t trust you with my life either. Are you useless then?
      - chx 3y ago
        Instead of standing up a strawman argument: why don't you trust it? How does that extend to other areas where you do ?
    - danielbln 3y ago
      If you use an LLM as a mere fact dispenser, then yes, they're quite susceptible to hallucinations. That said, there are search enabled LLMs, there is RAG to inject grounding truth into the LLMs context, there are LLMs that use their own code execution environments to validate code and output, all of which severely reduce hallucinations. Also, I would not rely on a single (online) source for matters of life and death (medicine, finding non-poisnonous mushrooms, etc.) so why would LLM output be any different? LLMs are not knowledge bases by themselves, they are text/language transformers, treat them as such or augment them with additional knowledge. Or don't, if you think millions of people are gaslighting you and LLMs are just a parlor trick, then I don't know what to tell you.
    - rashkov 3y ago
      > It's probabilistic so every answer it can give can only resemble the real one and only accidentally be the real one so you need to fact check it anyways and then why bother In the case of code or other computer knowledge, it’s easy enough to check that it works. Your same criticisms also apply to stack overflow, or any webpage you find through a search engine. Why bother when the content might be wrong or made up? What makes you trust those websites more than GPT?
    - rizky05 3y ago
      [dead]
hanselot 3y ago
I hate posting my stupid things here, because they are hardly ever relevant, but w/e here goes. Last night I spent a good couple of hours with ChatGPT assessing the similarities between DNA and Blockchain. Blockchain does seem to have quite a few similar features to what DNA. With smart contract type code you introduce a mechanism that can act like an immune system if programmed to, setting up bounties for proactive repair on the open market. I don't know why I want blockchain to act like synthetic DNA, but in my mind these things are more similar than distinct. As is the idea that blockchain could be the natural place to house AI. Why burn the planet down to mine hashes, when you could be mining inference. Anyways, I agree with your sentiment though. A lot of AI is the new crypto is going around, the only difference being, AI has way more potential upside (unless people realise the economy is dead, and only held aloft by trillions of dollars of paper that has next to no value once people realise how diluted it is). Ironically, I think we should listen to 4chan and buy silver xD

chx 3y ago

I am not participating in industrial strength gaslighting. "greatly improving what I can accomplish" is a trap. But to each their own.

danielbln 3y ago
You may be gaslighting yourself if you can't see the actual, tractable, current practical applications of this tech (and that includes local models you can run yourself).
- chx 3y ago
  That's the thing: there is no actual and tractable use of this tech. It's probabilistic so every answer it can give can only resemble the real one and only accidentally be the real one so you need to fact check it anyways and then why bother? Do consider the AI written mushroom hunting guides being sold on Amazon. Would you trust these literally with your life?
  - flappyeagle 3y ago
    No but I don’t trust you with my life either. Are you useless then?
    - chx 3y ago
      Instead of standing up a strawman argument: why don't you trust it? How does that extend to other areas where you do ?
  - danielbln 3y ago
    If you use an LLM as a mere fact dispenser, then yes, they're quite susceptible to hallucinations. That said, there are search enabled LLMs, there is RAG to inject grounding truth into the LLMs context, there are LLMs that use their own code execution environments to validate code and output, all of which severely reduce hallucinations. Also, I would not rely on a single (online) source for matters of life and death (medicine, finding non-poisnonous mushrooms, etc.) so why would LLM output be any different? LLMs are not knowledge bases by themselves, they are text/language transformers, treat them as such or augment them with additional knowledge. Or don't, if you think millions of people are gaslighting you and LLMs are just a parlor trick, then I don't know what to tell you.
  - rashkov 3y ago
    > It's probabilistic so every answer it can give can only resemble the real one and only accidentally be the real one so you need to fact check it anyways and then why bother In the case of code or other computer knowledge, it’s easy enough to check that it works. Your same criticisms also apply to stack overflow, or any webpage you find through a search engine. Why bother when the content might be wrong or made up? What makes you trust those websites more than GPT?
  - rizky05 3y ago
    [dead]

hanselot 3y ago

I hate posting my stupid things here, because they are hardly ever relevant, but w/e here goes. Last night I spent a good couple of hours with ChatGPT assessing the similarities between DNA and Blockchain. Blockchain does seem to have quite a few similar features to what DNA. With smart contract type code you introduce a mechanism that can act like an immune system if programmed to, setting up bounties for proactive repair on the open market. I don't know why I want blockchain to act like synthetic DNA, but in my mind these things are more similar than distinct. As is the idea that blockchain could be the natural place to house AI. Why burn the planet down to mine hashes, when you could be mining inference. Anyways, I agree with your sentiment though. A lot of AI is the new crypto is going around, the only difference being, AI has way more potential upside (unless people realise the economy is dead, and only held aloft by trillions of dollars of paper that has next to no value once people realise how diluted it is). Ironically, I think we should listen to 4chan and buy silver xD