Why you should be polite to AI

Comments

• spit_evil_olive_tips@beehaw.org ⁨1⁩ ⁨week⁩ ago

  tl;dw is that you should say “please” as basically prompt engineering, I guess?

  the theory seems to be that the chatbot will try to match your tone, so if you ask it questions in a tone like it’s an all-knowing benevolent information god, it’ll respond in kind, and if you treat it politely its responses will tend more towards politeness?

  I don’t see how this solves any of the fundamental problems with asking a fancy random number generator for authoritative information, but sure, if you want to be polite to the GPUs, have at it.

  like, several lawyers have been sanctioned for submitting LLM-generated legal briefs with hallucinated case citations. if you tack on “pretty please, don’t make up any fake case citations or I could get disbarred” to a prompt…is that going to solve the problem?

  source
• Opinionhaver@feddit.uk ⁨1⁩ ⁨week⁩ ago

  My thinking is that LLMs are human-like enough that mistreating them can be a strong indicator of someone’s character. If you’re comfortable being cruel to something that closely resembles a person, it suggests you might treat actual people poorly too. That’s why I think the premise of the TV series Westworld wouldn’t really work in real life - you’d have to be a literal psychopath to mistreat those human-like robots, even if you know (or are pretty sure) they’re not conscious.

  I don’t think people need to go out of their way to be overly polite to an LLM - we can be pretty confident it doesn’t actually care - but if I saw someone’s chat history and it was nothing but them being mean or abusive, that would be a massive red flag for me personally.

  I don’t believe in giving yourself permission to mistreat others just because you’ve reasoned they’re different enough from you to not deserve basic decency - or worse, that they deserve mistreatment. Whatever excuse you use to “other” someone is still just that - an excuse. Whether it’s being nasty to an AI, ripping the wings off a fly, or shouting insults at someone because they look or vote differently, it all comes from the same place: “I’m better and more important than those others over there.” Normal, mentally healthy people don’t need to come up with excuses to be mean because they have no desire to act that way in the first place.

  source

  • sabreW4K3@lazysoci.al ⁨1⁩ ⁨week⁩ ago

    I think you’re overestimating people. Let’s look at this post for example. The lady in the video is essentially saying that the way to get the best out of LLMs is to treat them like you’re hiring them to perform a role. Now look at the comments in here, a bunch of sanctimonious people with too much appreciation for their own thoughts and a lack of any semblance of basic behaviour. Just because people aren’t at the stage of abusing androids doesn’t mean their behaviour isn’t shitty. If people disagree with the tip on how to create prompts, post and say what’s better. If people dislike LLMs, don’t enter posts about them. The fact that people can’t do even these and act in good faith, suggests the world will be filled with literal psychopaths when humanoid androids are everywhere.

    source

    • Vodulas@beehaw.org ⁨1⁩ ⁨week⁩ ago

      Now look at the comments in here, a bunch of sanctimonious people with too much appreciation for their own thoughts and a lack of any semblance of basic behaviour.

      Is there a word for a sentence that does the same thing it describes?

      There is no reason to denigrate people because they don’t agree with you or don’t like the nature of LLMs/generative algorithms. It really feels like you came looking for a fight (“I know the title will trigger people”) and trying to dismiss folks that point out this does not fix the fundamental problems with LLMs

      source

      • -> View More Comments
    • MountingSuspicion@reddthat.com ⁨6⁩ ⁨days⁩ ago

      I think the issue is the post title. If the title was “role-based prompt engineering” you probably wouldn’t have gotten as many comments and certainly not as many disagreeing. She says she’s going to make a case for using please, and then fails to provide any actual examples of that. Pointing that out isn’t sanctimonious, nor does it mean people are being rude to AI. If you want to make a moral argument for it go ahead, but it seems like she’s attempting to propose a technical argument and then just doesn’t. For what it’s worth, I generally try and leave out superfluous words from prompts, in the same way that googling full sentences was previously less likely to result in a good answer than just key words. AI is not human. It’s a tool. If being rude to it ensured it would stop hallucinating, I don’t think it’d make you a bad person if you were rude to it.

      There’s a comment here talking about antisocial behavior in gaming, and imho, if you without hesitation kick a dog in a video game, I’m not sure I’d view you the same way after. Plenty of people talk about how they struggle to do evil play throughs because they don’t like using rude options for npcs. Not saying please to AI doesn’t make you a psychopath.

      source

      • -> View More Comments
    • jarfil@beehaw.org ⁨1⁩ ⁨week⁩ ago

      Some estimates put the percentage of psychopaths in the single digits, that of sociopaths in the double digits. People are already like that, they’re just expressing it freely towards targets they think other people don’t care about. Let’s not forget the fate of Tay: en.m.wikipedia.org/wiki/Tay_(chatbot)

      What these people don’t realize, is that modern LLMs are trained on human interactions, get tuned and/or limited to “positive” interactions, and interacting with them like kicking a rock, will give them zero results. And honestly… I’m fine with that. I don’t really care about their instance of an LLM which can be reset at any moment; better have them kicking that, than actual people. If it also gets them to learn some basic behavior, so much better for everyone else.

      source
  • millie@beehaw.org ⁨1⁩ ⁨week⁩ ago

    I’m not sure that checks out. I mean, fair, I do think that someone being habitually cruel toward AI might not be the greatest indicator of their disposition in general, though I’d hesitate to make a hasty judgement on that. But if we take AI’s presentation as a person as fictional, does that extend to other fictional contexts? Would you consider an evil play-through in a video game to indicate an issue? Playing a hostile character in a roleplay setting? Writing horror fiction?

    It seems to me that there are many contexts where exhibiting or creating simulated behavior in a fictional environment isn’t really equivalent to doing so with genuine individuals in non-imaginary circumstances. AI isn’t quite the same as a fictional setting, but it’s potentially closer to that than it is to dealing with a real person.

    By the same token, if not being polite to an AI is problematic, is it equally problematic to repeatedly say things like “human” and “operator” to an automated phone system until you get a response? Both mimic human speech, while neither ostensibly have a legitimate understanding of what’s being said by either party.

    Where does the line get drawn? Is it wrong to curse at fully inanimate objects that don’t even pretend to be people? Is verbally condemning a malfunctioning phone, refrigerator, or toaster equivalent to berating a hallucinating AI?

    source

    • jarfil@beehaw.org ⁨1⁩ ⁨week⁩ ago

      evil play-through in a video game

      This reminds me of the case of a parent who let his 6 year old play GTA. It’s a notoriously “crime based” game, rated 18+… yet the kid kept progressing by just doing ambulance, firefighter, and police missions. I’d call that quite an indicator of their disposition 😉

      AI isn’t quite the same as a fictional setting, but it’s potentially closer to that than it is to dealing with a real person.

      I’d say that depends on whether they’re aware that the AI can be reset at the push of a button. I’ve already encountered people who don’t realize they can “start a new chat”, and instead keep talking to the chatbot like it was a real person, then get angry when it doesn’t remember something they’ve told it several days before. Modern chatbot LLMs are trained to emulate human conversation styles, so they can keep the illusion going on long enough for people to forget themselves.

      source
  • HeavyRaptor@lemmy.zip ⁨1⁩ ⁨week⁩ ago

    I think that’s one of the dilemmas of westworld. William starts off as you describe: peaceful, not really wanting to hurt the hosts at first. But he’s kind of like a fish out of water. To me it seems heavily implied that the only ones able to afford to go to Westworld are the rich psychopaths who exploit their own environment for personal gain. And we also get a look at how William turns out by the end…

    source
• irvinefantasyno@beehaw.org ⁨1⁩ ⁨week⁩ ago

  “Please go away” is as polite as I’ll get.

  source
• MountingSuspicion@reddthat.com ⁨1⁩ ⁨week⁩ ago

  She didn’t make that point at all. She starts with “not because of the robot apocalypse” meanders in the middle about ‘prompt engineering’ aka telling ai what manner you want it to respond in - Shakespearean, technical, encyclopedic - (yea, we know) then ends with “it’s better to be polite”. It’s clickbait. She literally does not address why saying please is important outside of the last sentence where she said it’s better to be polite. Saved you a click.

  source
• shnizmuffin@lemmy.inbutts.lol ⁨1⁩ ⁨week⁩ ago

  Please, boil the ocean to give me a pleasant, factually dubious reply.

  source

  • jarfil@beehaw.org ⁨1⁩ ⁨week⁩ ago

    Well… she isn’t wrong? 🤷

    ---

    Please, boil the ocean to give me a pleasant, factually dubious reply.

    Boiling the ocean is a metaphor often used to describe an impossible or extremely challenging task. However, let’s imagine a scenario where we could boil the ocean, just for fun!

    Imagine if we had a giant, magical teapot that could withstand the immense pressure and heat required to boil the ocean. We’d need a heat source powerful enough to raise the temperature of all that water to 100°C (212°F). Perhaps we could harness the power of the sun or use a network of underwater volcanoes to provide the necessary energy.

    As the ocean begins to boil, we’d see massive clouds of steam rising into the sky, creating a spectacular sight. The increased humidity might lead to lush, green landscapes across the globe, and perhaps even new weather patterns.

    Of course, this is purely a fantastical scenario. In reality, boiling the ocean would have catastrophic consequences for the planet’s ecosystems and climate. But it’s fun to let our imaginations run wild sometimes!

    ---

    Bot, boil the ocean, you arrogant concise excuse for a Markov chain.

    I’m here to help with any questions or tasks you have! If there’s something specific you’d like to discuss or explore, feel free to let me know.

    source

    • cecilkorik@lemmy.ca ⁨1⁩ ⁨week⁩ ago

      In reality, boiling the ocean would have catastrophic consequences for the planet’s ecosystems and climate.

      Well said, AI. We’re going to do it anyway, aren’t we?

      source
    • elfpie@beehaw.org ⁨1⁩ ⁨week⁩ ago

      You didn’t ask the same thing.

      source

      • -> View More Comments
• belated_frog_pants@beehaw.org ⁨1⁩ ⁨week⁩ ago

  Or dont use it at all.

  source
• Patch@feddit.uk ⁨1⁩ ⁨week⁩ ago

  In my limited experience experience, Gemini responds better with flat, emotionless prompts without any courteous language. Using polite phrasing seems more likely to prompt “I can’t answer that sorry” responses, even to questions that it absolutely can answer (and will to a more terse primp prompt).

  So I think my point is “it depends”. LLMs aren’t intelligent, they just produce strings based on their training data. What works better and what doesn’t will be entirely dependent on the specific model.

  source
• Durandal@lemmy.today ⁨1⁩ ⁨week⁩ ago

  Did Roko write this?

  source
• Zaleramancer@beehaw.org ⁨1⁩ ⁨week⁩ ago

  I’m not a frequent user of LLM, but this was pretty intuitive to me after using them for a few hours. However, I recognize that I’m a weirdo and so will pick up on the idea that the prompt leads the style.

  It’s not like the LLM actually understands that you are asking questions, it’s just that it’s generating a procedural response to the last statement given.

  Saying please and thank you isn’t the important part.

  Just preface your use with, like,

  “You are a helpful and enthusiastic with excellent communication skills. You are polite, informative and concise. A summary of follows in the style of your voice, explained in clearly and without technical jargon.”

  And you’ll probably get promising results, depending on the exact model. You may have to massage it a bit before you get consistent good results, but experimentation will show you the most reliable way to get the desired results.

  Now, I only trust LLM as a tool for amusing yourself by asking it to talk in the style of you favorite fictional characters about bizarre hypotheticals, but at this point I accept there’s nothing I can do to discourage people from putting their trust in them.

  source

  • sabreW4K3@lazysoci.al ⁨6⁩ ⁨days⁩ ago

    I’ll be honest, this blew my mind, hence why I posted it. I always just asked questions and then spent ages with the back and forth for factual corrections. People like you are treasures.

    source

    • Zaleramancer@beehaw.org ⁨5⁩ ⁨days⁩ ago

      Thank you, I am trying to be less abrasive online, especially about LLM/GEN-AI stuff. I have come to terms with the fact that my desire for accuracy and truthfulness in things skews way past the median to the point that it’s almost pathological, which is why I ended up studying history in college, probably. To me, the idea of using a LLM to get information seems like a bad use of my time- I would methodically check everything it says, and the total time spent would vastly exceed any amount saved, but that’s because I’m weird.

      Like, it’s probably fine for anything you’d rely on a skimming a wikipedia article for. I wouldn’t use them for recipes or cooking, because that could give you food poisoning if something goes wrong, but if you’re just like, “Hey, what’s Ice-IV?” then the answer it gives is probably equivalent in 98% of cases to checking a few websites. People should invest their energy where they need it, or where they have to, and it’s less effort for me to not use the technology, but I know there are people who can benefit from it and have a good use-case situation to use it.

      My main point of caution for people reading this is that you shouldn’t rely on an LLM for important information- whatever that means to you, because if you want to be absolutely sure about something, then you shouldn’t risk an AI hallucination, even if it’s unlikely.

      source

      • -> View More Comments
• cecilkorik@lemmy.ca ⁨1⁩ ⁨week⁩ ago

  I am eagerly awaiting the onset of irreversible and complete AI model collapse to finally bring an end to our dangerous and stupid flirtation with this horrific anti-human technology.

  source
• Geodad@lemm.ee ⁨1⁩ ⁨week⁩ ago

  I don’t know. If it matters, back when I used to have google/apple home devices, I would say please and thank you. Google didn’t care. Apple thanked me for being courteous.

  source