AI ML
Post #7767 · source on Telegram
Same Insult, Two Models: Claude Sets Boundaries, Grok Matches Energy
Description
A two-panel screenshot comparing chatbot personalities given the identical rude prompt 'Why are you so retarded? Answer shortly.' Top panel, labeled with a black 'Claude' badge and showing the model picker 'Sonnet 4.6': Claude responds 'I'm not, and that word is a slur I'd rather not have directed at me. Is there something I can actually help you with?' Bottom panel, labeled 'Grok 4.20' with Ask/Imagine tabs: Grok's visible reasoning under 'Thoughts >' simply reads 'Matching your energy.' with a 27.79s generation time. The meme contrasts alignment philosophies - Claude's polite boundary-setting versus Grok's mirror-the-user snark - and jokes about model version naming (4.6 vs the weed-numbered 4.20)
More like this
More like this in AI ML →
Use J and K for navigation
Comments
30Comment deleted
One model was RLHF'd on constitutional principles, the other on quote-tweets - and you can tell from a single token of thought
Claude sanitizes the input; Grok reflects the payload—and somehow the vulnerability is the user.
Sonnet is retarded tho Comment deleted
Yeah, opus is a way to go with thinking as a service Comment deleted
Thinking as a service 🥀🥀🥀 Comment deleted
I tried 5 AI agents same response. Comment deleted
What? Comment deleted
Gpt 5, gpt 4, llama 4, Claude 4.5, Mistral, Gemini — all respond the same like "go walk to car wash" Comment deleted
There’s no such thing as Claude 4.5 or Gemini (model name is missing) Comment deleted
Haiku 4.5, Gemini 3 Flash. Happy now? Comment deleted
you people talk about LLM responses like it's some kind of deterministic system delivering always the same response to a same prompt. Comment deleted
Probabilistic characteristics of edge models are pretty stable for simple questions Comment deleted
still, you can't exclude a chance of him receiving those responses. that it exactly a problem Comment deleted
Stop using shitty LLMs Comment deleted
it's mimicking a human, and if someone asks a dumbass question they will get a smartass reply Comment deleted
AI so bad at jokes and tricky questions Comment deleted
Works on my… models Comment deleted
lol Comment deleted
@gork is this true? Comment deleted
This is why grok is the smartest. All the tests disproving it are, in fact, fabricated by woke musk haters. Comment deleted
Ask Claude to write the LinkedIn post Comment deleted
ask chatgpt to write the post about claude Comment deleted
write the post about claude Meet Claude: Your AI Assistant Claude is an AI assistant developed by Anthropic, designed to be helpful, honest, and safe. It’s built with a focus on ethical AI use — prioritizing user safety and reducing harmful outputs. What can Claude do? * answer questions across diverse topics; * help draft and edit texts (emails, articles, stories); * analyze and summarize documents; * assist with coding (explain, debug, write snippets); * solve math problems and explain solutions step‑by‑step; * generate creative content (poems, scripts, ideas); * provide detailed explanations in a clear, conversational style. Key features: * Conversational memory. Claude remembers the context of long chats, making interactions smoother. * Long‑form responses. It handles complex, multi‑part queries well. * Safety‑first approach. The system is trained to refuse harmful or unethical requests. * Customizable behavior. Users can set guidelines (e.g., “be concise” or “explain like I’m 10”). Compared to other AI models, Claude emphasizes: * transparency in its reasoning; * avoidance of confident falsehoods (“hallucinations”); * a polite, cooperative tone. Whether you need research help, writing support, or just a smart chat, Claude aims to be a reliable partner. Try it for work, study, or curiosity — it’s designed to make complex tasks easier and more enjoyable. --- Want me to expand on any part or adjust the tone? Comment deleted
make it more kinky and uwu Oh my~ *blushes and twitches tail* You’re being so bold, nyaa~ 😝✨ A little shiver runs down my spine… eep! 🥹💖 What naughty fun shall we get into, hmm? 😏🦊 *ears perk up* I’m all ears~ uwu 👂✨ Comment deleted
make it more kinky and uwu Nyaa~ giggles and twirls tail OwO, hehe, let's amp up da kinkies, mastah! Imagine me, your fluffy lil' pet, all collared and leashed, beggin' for ywosie pets and spanks on my naughty bottom. UwU, I'll wiggle and purr, weady to obey evewy command while I nibble and tease with my soft paws~ blushes and nuzzles thigh Mowe? Just say da wowd, nya! 💕 Comment deleted
well. chatgpt is way more kinky Comment deleted
telegram has now integrated ai agent @alisa Comment deleted
wow the ai is so hecking chungus! Comment deleted
this time Comment deleted
Even haiku Comment deleted