• 0 Posts
  • 312 Comments
Joined 3 years ago
cake
Cake day: June 11th, 2023

help-circle


  • I agree that it’s awkward and it creates problems in some scenarios. I’m not sure it’s necessarily a bad thing, but it’s an understandable (over) correction for some pretty heinous behaviour.

    I think it’s also equally pretty weird to just dismiss a facet of someone’s humanity.

    At work? I think it’s preferable to limit our interactions to work related stuff.

    Outside of work, I’m ambivalent. We associate physical features, names, and accents with cultures. But those don’t always line up to significant differences in someone’s personality.


  • In the last few decades we’ve noticed that we’ve been treating each other like shit. We’ve used race, skin colour, ethnicity, weight, etc to insult others and reduce their social standing.

    We’re trying to fix that. As such, calling out those specific differences is frowned on, even if we aren’t using them negatively.

    Is this inconvenient? Yes. It’s pretty easy to point out the only black/fat/disabled person in a work place. But we’re really trying to avoid any conversations that could turn into insults or attacks.

    So we now have an unwritten social rule that we avoid using those identifiers when talking about individuals.





  • Or the AI on shopping websites saying “I’d recommend this model…”

    We don’t have a pronoun for “this non-human unit”. LLMs are marketed as conversational, so they need to conform to the limitations of English.

    One could argue that “we” or “one” would be more appropriate, but that would sound stilted in many contexts.

    I’d prefer linguistic markers to distinguish between people and machines, but we haven’t gotten there yet.








  • Generally speaking, you need to use social signals: does it seem like other people are using the software? Is it recommended by people you trust? Does the author look legit (other projects, a presence on social media, etc)?

    That’s because it’s really easy to hide malware. Developers can’t read an entire codebase, and the codebase of every library required by the tool.

    In the ideal scenario, permissions on your home directory are configured appropriately so an attacker can’t do too much damage. I’m not sure if that’s realistic, however.

    There have been lots of stories about supply chain attacks that steal developer’s crypto wallets, which is a perfect illustration of the problem.

    Edit: running everything in a VM is probably the safest way to deal with untrusted code.