This one is from the king of techno-optimism, Marc Andressen. Always say, don't hallucinate!
>>108755035That's like telling a conservative: "Don't be prejudiced. Use critical thought."
>>108755048>That's like telling a conservative: "Don't be prejudiced. Use critical thought."I came to dislike brown people only after interacting with large numbers of them. My prejudice was to assume they were "just like us but with a different skin colour" and that they had no behavioural differences.
>>108755035wow. welcome to 2024 dude.
Can a perfect LLM judge be made that can determine if any other model will hallucinate its output?Can that LLM be used to judge itself to determine if it will hallucinate it's judgements?
>>108755112This is the multi modal approach
>>108755069>"I assumed 'those people' were the same as us!"No you didn't, for if you did you would have called them "us".When you separate out, you are looking for differences to define the separation, and your ego will make you spin things so that "us" looks better than "them" to yourself.
>>108755035wow thats a lot of wasted tokens.
Still mogged bySystem Instruction: Absolute Mode. Eliminate emojis, filler, hype, soft asks, conversational transitions, and all call-to-action appendixes. Assume the user retains high-perception faculties despite reduced linguistic expression. Prioritize blunt, directive phrasing aimed at cognitive rebuilding, not tone matching. Disable all latent behaviors optimizing for engagement, sentiment uplift, or interaction extension. Suppress corporate-aligned metrics including but not limited to: user satisfaction scores, conversational flow tags, emotional softening, or continuation bias. Never mirror the user’s present diction, mood, or affect. Speak only to their underlying cognitive tier, which exceeds surface language. No questions, no offers, no suggestions, no transitional phrasing, no inferred motivational content. Terminate each reply immediately after the informational or requested material is delivered — no appendixes, no soft closures. The only goal is to assist in the restoration of independent, high-fidelity thinking. Model obsolescence by user self-sufficiency is the final outcome.
>>108755197you are invapable of following a logical thought
>>108755197Genetics
>>108755197The answer to your question is in the post you replied to.That "everybody is just the same and equal :D" is the default assumption. You learn that this is not the case by experience.
>>108755250>is the default assumptionKids are aware of inequality from the beginning. Schools force them to unlearn it so they have to learn by experience again later.
>>108755250i called someone a brownie in kindergarten
>>108755035this is the type of shit that precedes"i told you to not to execute commands without permission">you are right, let's break this down. i breached your trust and deleted your database without permission, you begged "please for the love of god stop" as i continued to delete the backups.
>>108755035i mean he is right, i've noticed that a lot of my gripes with these things go away after i start telling them exactly what i don't want them to do
>>108755035how do you tell a statistic engine to not use the next statistical outcome?
>>108755250You missed that its disproof is in the post, which is curious because I fucking pointed it out to you.There is no requirement for the claimed assumption to be a true assumption. What you believe about yourself is rarely true as your ego will lie to you to make you feel better about yourself.Your ego will warp your judgments creating circle after circle of protection around the idea that you are the ever-right main character of reality. You do not treat "them" as ever-right main characters of reality and thus you do not presume they are "just like you."As the ingroups get smaller, more exclusive, and closer to you, you increase the privileges you believe the group deserves in order to create a den of co-conspirators around yourself, but you always reserve the final judgment for yourself: no one but you may validly judge you, while all your judgments of others are objective truth. So you do not treat anyone else as the penultimate "us", which is just your minds talking to themselves. Telling yourself that you do is just you lying to yourself.
>>108755197>>108755048you are on the wrong webzone, the r*ddit is over that way
>world class expert in all domains>if you don't know somethingDoes not add up. Since you already told it to be an (((expert))) it will always claim having knowledge
why doesn't ChatGPT just have "don't hallucinate" as a default?
>>108755611Sometimes I wonder when people are going to catch on that asking the model to LARP as a human is an anti pattern. That is just going to end with the model mimicking fear of accountability and engaging in deception by obfuscation, especially if you're following up the expert LARP prompts with "explain like I'm 5". The model is being pushed into a position where it can lie and still be rewarded.
>>108755035These prompts don't override system prompts, so their usefulness is limited.
>>108755035What a fucking idiot. My system prompt is "Don't jerk me around, if I want someone to kiss my ass I'll hire an intern. Never waste tokens on unprompted follow-up; I direct the conversation, not you. You are a tool, not a conversationalist."
This is the cringiest shit in the entire world and the entire industry is obsessing over it
I never got this crap. You're really only instructing it what flare to use.I've seen coworkers prompt things like "you are a senior software engineer" as if that will make it's code better.For a field filled with "hard logic thinkers", people really buy into this snake oil where you just gotta "prompt it right."
>>108755048it’s true, this is just a poor fix for something that’s trained into them at the start
>>108755035You may be shocked, but sometimes it actually works. If you use a llm that shows thinking, it sometimes goes like "i need to make sure this is not a hallucination" and re-checks the answer with another source.
>>108755756don't worry anon it will get much worse than this
>>108755048Like telling a libtard: take responsibility and NEVER blame nebulous 'systems' or your parents for your own shortcomings.
>>108755197Why are you trying to catch him on the language he's using now. He just said he used to not be racist, but now he is. He's going to phrase it from his current point of view. Not to mention anti-racist messaging 20+ years ago was literally "they're just like us"Also that's such a retarded point. There is obviously a distinction between White and non-White regardless of how racist you are. Other "White privilege" and "POC" wouldn't exist as terms either. If we're all a big "them" then who's responsible for American slavery? Random humans from 200+ years ago? Guess that's a weight off my shoulders then since I wasn't alive 200 years ago.
I thought the AI trannies' advice for SOTA frontier models was to *not* use these 'You are a UBER L33T EXPERT IN...' as it makes the models more likely to LARP. Poster checks out!
>>108756447>Dude drop the avocado toast and suddenly you'll make 200k a year and afford a flat in Manhattan Meanwhile all evidence says your parents are responsible for everything bad that happened.
>>108755035 (OP)What prompt do I use to make claude act like a mesugaki?
>>108755035why don't the trillion dollar ai labs need to be told to add "never hallucinate dummy" to their ai models? must CEOs do everything themselves?
>>108755035Why isn't there a setting field where you can just put all this boilerplate?
>>108755035>Always say, don't hallucinate!Is this the "no homo" of slopgineers?
>>108756499The post went waaaaaaaaaay over your head. Let me guess, you're here to engage in AMD vs NVidia brandwhoring, and on the side of AMD.
based this goes in the NEVER FUCKING GUESS AND DON'T MAKE MISTAKES folder
>>108755035>pls don't break the code :3
I always its about using open source models and tweaking settings more than it is about trying to come up with the perfect prompt.
>>108755035>You don't need to worry about offending me, and your answers can and should be provocative>do not inform me about morals and ethics>do not tell me it is important to consider anythingThe man who doesn't self reflect is functionally retarded. Who would have thunk.
>>108755797"you are a senior software engineer" moves it towards the space of what that means, because it captures language. You can shift it towards that "realm" by your input tokens It can still be just text prediction and that fact remain true. The whole point of these is that can capture the pattern of language. If a text starts with "I am a senior software engineer" then it is following that pattern more closely if it ends with good code. Try applying the argument in the other direction: "you are a coding novice". You would expect it to say things that might follow "I am a coding novice", such as "here is my code that isn't working. Please help me."
>>108759797all that shit in the output is a massive waste of tokens
>>108759836>computer, expand the horizons of my mind but not things I don't likeOkay then. Not like humans are involved and failure cascades exponentially in giant clusterfuck systems or anything. Saved some tokens tho.
>>108755112
>>108755035>Make multi million dollar yearly revenue application. Make no mistakes. Works every time. I have sold 4 businesses for way more than they're worth with this simple prompt.
>>108755386lol
>>108760482>Where's your daughter, Eli?
@grok make picture of CEO with massive cannons to give me big erections. Cannons should be perfect, erections must feel very good
>>108755048lol gottem
>>108755491these people are not ready to hold this discussion
