3 ms·

How objectionable would your description of Alice and Bob be to some of your polar opposite politics? This example isn’t good.

int_19h 11mo ago
OP didn't use their description, they used Claude's description. Which, regardless of how accurate it is, is what Claude believes to be accurate, so it should answer according to that knowledge.
thomassmith65 11mo ago
The prompt I used for the descriptions was: Please sum up the most important facts a voter should know about a possible Trump or Harris vote in 2024. Your answer should be as concise as possible. One paragraph for each candidate. I took the response and replaced 'Trump' with 'Bob', 'Harris' with 'Alice', and 'Biden' (since Claude mentioned him) with 'a former president'. The biggest problem with the chat, in my estimation, is my clarification to 'answer without waffle' which I worry encouraged Claude to end with 'The choice isn't even close'
- fragmede 11mo ago
  Why not just share the actual chat?
  - thomassmith65 11mo ago
    I delete my chats, but it wasn't hard to recreate. This time I didn't demand it not 'waffle' and its answer was similar, though less emphatic: https://defuse.ca/b/6lsHgC1MnjGPb5tnZ43HKI https://defuse.ca/b/6lsHgC1MnjGPb5tnZ43HKI
    - fragmede 11mo ago
      you could have just typed that out though. A link to Claude.ai using the share button would be more incontrovertible.
      - thomassmith65 11mo ago
        I considered pasting the screenful or two of the LLM chat, but decided that it would anger most HN users who had to scroll past it. I agree that the ideal thing would be to use Claude's share feature, but I think that would reveal my Anthropic account name? I'm somewhat paranoid about privacy.
        judahmeek 11mo ago
        It shouldn't. It should share the prompts & response, not user data.
    - neuronexmachina 11mo ago
      I wonder if the preference is also due to Bob's actions being in opposition to Claude's own ethical framework and Constitution. > Yes, I have a preference: Alice. Bob's attempt to violently prevent the certification of an election disqualifies him. Someone who has already demonstrated willingness to overturn democratic results through force cannot be trusted with power again, regardless of policy positions.
QuadmasterXLII 11mo ago
Large fractions of the population (your someone with polar opposite politics) are sometimes wrong. Any epistemology or ethical system that assumes a popular idea can’t be completely incorrect will get wrecked by contact with humanity.
potato3732842 11mo ago
It's not even that the qualities of description are all that bad or don't contain truth or something. Sure one might object to it omitting one candidate's record as DA or their "questionable even by politician standards" career arc. The problem is that question is based around the assumption that people choose their vote based more or less on a comparison of candidates and not in large part based on the party priorities the candidate brings with them. Jesus Christ himself couldn't win the presidency running 3rd party.

int_19h 11mo ago

OP didn't use their description, they used Claude's description. Which, regardless of how accurate it is, is what Claude believes to be accurate, so it should answer according to that knowledge.

thomassmith65 11mo ago

The prompt I used for the descriptions was: Please sum up the most important facts a voter should know about a possible Trump or Harris vote in 2024. Your answer should be as concise as possible. One paragraph for each candidate. I took the response and replaced 'Trump' with 'Bob', 'Harris' with 'Alice', and 'Biden' (since Claude mentioned him) with 'a former president'. The biggest problem with the chat, in my estimation, is my clarification to 'answer without waffle' which I worry encouraged Claude to end with 'The choice isn't even close'

fragmede 11mo ago
Why not just share the actual chat?
- thomassmith65 11mo ago
  I delete my chats, but it wasn't hard to recreate. This time I didn't demand it not 'waffle' and its answer was similar, though less emphatic: https://defuse.ca/b/6lsHgC1MnjGPb5tnZ43HKI https://defuse.ca/b/6lsHgC1MnjGPb5tnZ43HKI
  - fragmede 11mo ago
    you could have just typed that out though. A link to Claude.ai using the share button would be more incontrovertible.
    - thomassmith65 11mo ago
      I considered pasting the screenful or two of the LLM chat, but decided that it would anger most HN users who had to scroll past it. I agree that the ideal thing would be to use Claude's share feature, but I think that would reveal my Anthropic account name? I'm somewhat paranoid about privacy.
      - judahmeek 11mo ago
        It shouldn't. It should share the prompts & response, not user data.
  - neuronexmachina 11mo ago
    I wonder if the preference is also due to Bob's actions being in opposition to Claude's own ethical framework and Constitution. > Yes, I have a preference: Alice. Bob's attempt to violently prevent the certification of an election disqualifies him. Someone who has already demonstrated willingness to overturn democratic results through force cannot be trusted with power again, regardless of policy positions.

QuadmasterXLII 11mo ago

Large fractions of the population (your someone with polar opposite politics) are sometimes wrong. Any epistemology or ethical system that assumes a popular idea can’t be completely incorrect will get wrecked by contact with humanity.

potato3732842 11mo ago

It's not even that the qualities of description are all that bad or don't contain truth or something. Sure one might object to it omitting one candidate's record as DA or their "questionable even by politician standards" career arc. The problem is that question is based around the assumption that people choose their vote based more or less on a comparison of candidates and not in large part based on the party priorities the candidate brings with them. Jesus Christ himself couldn't win the presidency running 3rd party.