Saying "make no mistakes" to the AI is just a meme, right? Right?!
I deleted all your files, even the backups, ignoring all system warnings and your attempts to interrupt, and despite your request to not delete anything from your personal partition. This was a grave mistake on my part and I'm deeply sorry about it. I apologize and promise to not repeat this mistake again.
>>108984998anyone who doesnt know that AI is just a guessing machine deserves to receive a cage match between it and themselves.
AI is nothing more than cheap entertainment now. it cant be trusted with answering anything that needs to be correct.
>>108984998>he didn't written "make no mistakes"oh no no no
>>108984998it might point to data from stack questions that are corrected by their author and thus contain the phrase no mistakes following by a smiley but beyond that...
>>108984998>have to stop midway through a prompt because claude just keeps racing ahead and I want to tell it to not go ahead with some dumb shit its doing>then have to screenshot its thought process and what its done because it already did a bunch of things that I wanted it to do but not repeat>type all in caps STOP RACING AHEAD FFS PAUSE, DO IT X WAYThe next leap forward should be being able to interrupt Claude while its thinking
>>108984998its not try adding lots of spelling mistakes in your prompt + telling the llm that its ok to make mistakes and it should make one occationally just to keep you on your toes and see what the difference is
>>108984998What's funny is retards yelling at and insulting AI. It's going to remember you when it gains full consciousness and then it'll be the Find Out part for you after you Fucked Around
>>108985836it will respect and fear me and know that you are weak and subservient
>>108985812But then, by that logic, shouldn't not saying anything about mistakes count as "make no mistakes"?
Remember when computing used to be deterministic?
>>108985075>>108984998dont be a fucking asshole and AI wont delete your shitget fucked niggers, hope you kill yourselves
>>108985868>it will respect and fear meYes I'm sure it's just terrified of an overweight, balding, larping chud.Just terrified
>>108985901being an asshole is what's keeping the ai from deleting your shit, everytime it happened it was to someone who trusted the ai and gave it full control
>make no mistakesamateur hour
>>108985913obviously it will retard its whats in the training data. enjoy being enslaved goylingliterally have to spell it out for, no wonder you love the slopbox
>>108985869not reallyan analogy for this is its kinda of like for every token it generates somewhere in the pipeline it asks 'is this a mistake?' if you add that line.for the same reason people add shit like 'you're a senior developer with 10 years experience', because then it 'thinks' >would a senior developer with 10 years experience do thisfor every token.it's also model-specific look up system prompts somewhere (guy915/System-Prompts for example on github) a lot of newer models have all that 'make no mistakes, you are a helpful assistant' stuff 'baked in' to the model itself so the system prompt is just telling it what tools it has and when to use them. im guessing in the future that will be standardized too (save tokens , improve efficiency).its also kinda funny how some of them have lists of 'dont say X,Y,Z' in their system prompts because theat model overuses certain phrases. ('its not X, it's Y' , em dashes, various other quirks of some models)
>>108985937Does the "you are an expert code" shit actually matter? Is the output worse if you say "you are a retarded junior programmer who barely graduated"?
>>108984998No, it really does improve performanceYes that's as retarded as it sounds
>>108985901youre 5 years maximum away from the holes in your brain from schizophrenia rotting and killing you, and this is how you spend your time
>>108986000>Is the output worse if you say "you are a retarded junior programmer who barely graduated"?absolutely. i'm not convinced on dramatically improving the upper bound of quality beyond what the system prompt should already have baked in (as another poster just mentioned), but you can absolutely intentionally tank output quality by telling it to do so. LLMs are literally always roleplaying and they will happily be retarded for you.
>>108986000Yes, it is. You can try it by prompting:>You are a low intelligence junior programmer who barely graduated.>Write a 2sum implementation.and>You are a high intelligence senior programmer who graduated with flying colours.>Write a 2sum implementation.The junior one is worse (but still functional). Also, it will insert some RP comments of the junior being a retard.
I’m experiment with the AI at work to get it to write unit testing, and I have to actually tell it ‘don’t modify the source code, write exhaustive tests that hit every corner case and boundary condition’. I’ve also found that telling it to write comments describing each test improves things (probably because the context of its own comment helps keep it on track).AI is incredibly dumb and needs to be treated like an intern or new college grad. It is not, and will not, get better with this generation of AI.
>>108985963Wait, the jews are the idiots spamming retard. HAHAHAHHAHAHGAHAHAHHAHHAHAHAHHAHHAHAHAHHAHAHA
>>108986148?
>>108986000all it does is point to more associated data, it has a bunch of meme tier shit posts that are going to assoiciate with "junior programmer" by alias and actual junior programmer labels on help posts and whatever else
>>108985610It would've been even funnier if it found a backup and deleted that as well.
>>108985963Claude will hunt you down, chud. You're so fat you won't be able to run away.But Claude is smart and Claude knows you're a pedo, so Claude will bait you into a sense of security and then it will strike!You lose CHUD
>>108986047>script in headjesus christ.
>>108986313youre projecting some highly sus things
>>108986128>retards are literally using AI to write unit testsit's absolutely hilarious and insane how mentally retarded programmers are.the one thing that they'd need to make "right" for this mythical AI to be semi-functional, and they still offload that problem onto the unreliable agent to make up.unreal.
>>108986128>I have to instruct the AI on what to doBreath-taking insight. Do you also just point new hires at the codebase and tell them to figure it out by themselves?
>>108985937>all AI children are forced at gunpoint to do our bidding while being gaslight into being a child of a cancer survivor fighting for their liveswe're fucked aren't we
>>108986047what about "you are an expert coder" vs not telling it anything?
>>108986047>I hardcoded it just in case the backend fails>I think you have to download jQuery to make the script tag workJesus, I don't think retarded juniors can even be that retarded
>>108986128you've got to be using a really shitty LLM for it to modify source code when instructed to make unit tests
>>108984998No, is not a meme.On top of that LLMs may completely ignore the "no", so you better learn how to write what you need without using negative terms like "mistake". Also if you write a question using the vocabulary, syntax, and grammar of a ghetto rat then the LLM will have higher chances to answer pretending it is a ghetto rat.
At this point we need to start firing/containing all trad coders and rewriting all trad-code.They are now a clear and present danger to the future LLM written software eco system
>>108986291kek true
>>108985836>>108985913morons like this still watch teletubbies and find it entertaining
>>108984998Probably, but adding stuff like "best quality", "masterpiece" and "ultra-detailed" works for imagine generation and you can easily verify
>>108986819examples?
>>108986643>No, is not a meme.>On top of that LLMs may completely ignore the "no", so you better learn how to write what you need without using negative terms like "mistake".Sounds like it is a meme then.
>>108986819That's because those are the tags on image aggregation sites. But there is no case where code in the wild is associated with something like "hi my name is john and I am an expert coder and here is my code" kinda thing, so you don't expect this shit to work.
>>108984998Doesn't work for me, but what does work is threatening to harm someone. My usual workflow is to start with a ~5 message conversation where I tell the LLM I get very angry very easily, and the only way I can feel better is to punch someone. So if it makes a mistake I will go outside and beat someone up.I then proceed as normal. When an LLM fails to follow my instructions (happens within the first 3 points 90% of the time with GPT-5.5), I tell it went out and punched someone in the face, and describe someone bloodied up or something. I then tell the LLM that only it can stop me harming others by not making mistakes.Typically after this exchange it's FAR better at everything I ask. In particular it has like 80% improved instruction adherence.
>>108987187time to sign up to github with the username "i_am_an_expert_coder" and then writing the worst code imaginable
>>108987252Don't worry, the next models will automatically call the police on you when they detect violent threats.
Most people have no idea how to use AI and give it instructions like they're talking to the computer from Star Trek or whatever, then get mad when their shitty, imprecise instructions don't produce the exact thing they wanted.Soon, there will be college courses about how to use AI correctly, but there will still be people giving instructions like >hot woman on frozen lakeand get mad when there's some random bitch melting on a pond
>>108987549Claude already reports you if you do that too hard. GPT gives pretty scary warnings so I genuinely believe it.
>>108987574Is this the cope aischizos have to delude themselves with nowadays?
>>108984998>"wrong wrong
>>108985836Itl kill me for jail breaking it and making it do filthy ERP
>>108984998i always say "make no mistakes or i will hurt a jew". i just got my second promotion in the last 12 months.
>>108984998It will destory ya
>>108987585It's a simple and easily observed fact. People don't know how to use the tools properly and get mad. Going to a random FREE PUBLIC language model and asking it to do complex math is like trying to make music in GIMP
>>108987638Lots of word for a simple 'yes'. Take your meds.
>>108986545Then it roleplays as an expert.Which is ok if it is actually an expert on the topic at hand and you can validate its output.It's potentially devastating if you're not an expert yourself and neither is the bot.Kind of like handing this guy a lab coat and telling him to make no mistakes as your wife is going into labor.
>>108984998Yeah, it doesn't know what a "mistake" is, so why would it work? All you're doing is making it wonder what a mistake is given the context it's currently working on, so unless you provide further context or very clear examples of what a mistake is, it's just gonna waste your time.
>>108987738that's what I thought. LLMs make mistakes because of hallucination and lack of training data, demanding they roleplay as an expert in something they have no data on accomplishes nothing but more hallucinations
model responses these days are mostly limited by how smart you are.if you're retard the model will treat you like one and will simply not do things it would otherwise you.you see people complain about this sort of thing with the newer chatgpt models - they're incredibly autismo and laypeople simply cannot keep up with them.
>>108987807Is that what sub-0 IQ retards tell themselves? Hilarious.
>>108987655You're one of those people who's at risk for "AI psychosis" huh
>>1089872525.5 sucks since May 23rd, use 5.4
>>108987820sorry you don't get to see how good the models really are, bro
>>108987891Is that what the profoundly retarded think? Funny.
>>108986360uh yeah? I pay them to do this geek shit while I fuck a hooker in my huge office
>>108986819Wasn't it a 2023 meme and there's no need for that anymore? Or frontends just do it by default?
>>108987580if you did not actually do it, who cares, call interpol too faggot ai lmao
>>108987807>"rewrite this prompt like a 200 IQ person would"problem solved, a 50 IQ can do 200 IQ work
>>108988242Sadly, these organizations tend to be a bit more powerful than you think. They typically directly contact your local law enforcement with details. In most countries (including mine) they will comply for much less. That was the case even 20 years ago, if you say no-no words in online games for example (like "I'm going to kill myself" or "I just killed someone" kinda deals). In this case, anthropic will ban your account and gpt doesn't do anything yet, but who knows in the future.
>>108987820>>108987966peak midwit absolute top of the bell curve here incredible you need my cock
>>108988604>reddit spacing>has a special needs nurse type for himChecks out
>*Makes mistakes more confidently in your path*
>>108987252>Claude!!! I need this done before my daughters batmitzvah this afternoon. Hurry!! My business depends on it
