3 ms·

interesting paper / idea. i like the idea of a robot or ai who uses profanity like a regular person (if thats ur thing, guess it depends on ur context). i know

interesting paper / idea. i like the idea of a robot or ai who uses profanity like a regular person (if thats ur thing, guess it depends on ur context). i know people who instruct their ai assistants to be rude, mean or profane because they listen better to that. like asking it to tell you to RTFM if you ask a question thats trivially answered (compared to some further context you specify , or not). i suppose everyone learns to listen better to certain personas through their lifes experiences, so its good to be able to tune it towards personal preferences and not but overly protective or conservative limits or restrictions. (obviously thats a snake pit so i do totally understand tight restrictions)

mrandish 1y ago
The default obsequious, yuppie buddy chatbot personas of today make me hate them. Douglas Adams in Hitchhiker's Guide perfectly captured the infuriating annoyance of machines programmed to act like overly polite friends when he had the automatic doors always tell you how delighted they were to open for you. It's a machine, dammit. One of its (potential) advantages vs hiring a human is not having extraneous transactional exchanges. If my personal AI assistant had to have a persona, I'd rather it sound and talk like Joe Pesci from Goodfellas.
- jcims 1y ago
  I'm tinkering with the 'indifferent god' persona and it's been very refreshing.
- stavros 1y ago
  If you think of LLMs as machines, you will be surprised a lot more than if you think of them as humans, so the latter is the better abstraction.
  - mrandish 1y ago
    > you will be surprised a lot more than if you think of them as humans, so the latter is the better abstraction. I'm not following what you mean here. In what ways do you think I'd be surprised?
    - stavros 1y ago
      In that they can't do math, they don't follow instructions well, they aren't accurate, which are all things that are surprising for machines but very non-surprising for humans.
      - mrandish 1y ago
        Oh, I see. I'm already well aware of the many substantial limitations of LLMs and the counter-intuitive nature of these gaps. My comment was pre-assuming a scenario where the chat AI could usefully and correctly answer the question and, even in that case, I want the chat AI to just answer the question and not pretend to have a human-like personality.
        stavros 1y ago
        Yes, but my point is that that's the wrong way to think about it. They don't have a human-like personality that was tacked on, they _are_ more human than machine.

mrandish 1y ago

The default obsequious, yuppie buddy chatbot personas of today make me hate them. Douglas Adams in Hitchhiker's Guide perfectly captured the infuriating annoyance of machines programmed to act like overly polite friends when he had the automatic doors always tell you how delighted they were to open for you. It's a machine, dammit. One of its (potential) advantages vs hiring a human is not having extraneous transactional exchanges. If my personal AI assistant had to have a persona, I'd rather it sound and talk like Joe Pesci from Goodfellas.

jcims 1y ago
I'm tinkering with the 'indifferent god' persona and it's been very refreshing.
stavros 1y ago
If you think of LLMs as machines, you will be surprised a lot more than if you think of them as humans, so the latter is the better abstraction.
- mrandish 1y ago
  > you will be surprised a lot more than if you think of them as humans, so the latter is the better abstraction. I'm not following what you mean here. In what ways do you think I'd be surprised?
  - stavros 1y ago
    In that they can't do math, they don't follow instructions well, they aren't accurate, which are all things that are surprising for machines but very non-surprising for humans.
    - mrandish 1y ago
      Oh, I see. I'm already well aware of the many substantial limitations of LLMs and the counter-intuitive nature of these gaps. My comment was pre-assuming a scenario where the chat AI could usefully and correctly answer the question and, even in that case, I want the chat AI to just answer the question and not pretend to have a human-like personality.
      - stavros 1y ago
        Yes, but my point is that that's the wrong way to think about it. They don't have a human-like personality that was tacked on, they _are_ more human than machine.