The Evolution of Childhood Bragging Rights, from Dads to AI Agents
Description
A two-panel meme format comparing childhood arguments of the past to the present. The top panel, labeled 'THEN', shows two young boys facing off. One boy says, 'My dad is stronger,' to which the other replies, 'Oh yeah well my dad is smarter.' The bottom panel, labeled 'NOW', depicts the same two boys. The first boy boasts, 'My agent is natively multimodal and context aware,' using modern AI jargon. The second boy retorts with a seemingly simple but tricky challenge: 'Oh yeah well my agent knows how many R's are in "strawberry".' This meme humorously contrasts the straightforward boasts of the past with the complex, buzzword-laden claims of today's technology. The technical joke lies in the fact that while high-level AI concepts like multimodality are impressive, Large Language Models (LLMs) have historically struggled with deceptively simple, specific tasks like counting characters in a word. It's a satire on AI hype versus actual, sometimes surprising, limitations
Comments
17Comment deleted
One agent has a 1.8 trillion parameter model that's natively multimodal; the other can pass the 'strawberry' test. Let's be honest, the second one is the real AGI
We’ve gone from arguing about CPU clock speed to arguing whether your 175-B model can count the R’s in ‘strawberry’ - apparently Moore’s Law now measures ego per token
The real flex isn't having an agent with RAG, function calling, and chain-of-thought reasoning - it's having one that can finally pass the kindergarten spelling test that stumped GPT-4 for months
The strawberry test became the 'FizzBuzz of LLMs' - a deceptively simple problem that exposed how models trained on trillions of tokens could still fail at counting three R's. It's the AI equivalent of discovering your distributed system can handle petabytes of data but crashes when someone sends a null byte. Turns out 'natively multimodal and context aware' doesn't include 'can count letters in a word,' which is roughly the AI version of bragging about your microservices architecture while your health check endpoint returns 500
Before we ship anything "agentic and multimodal," can it pass the strawberry‑R eval at T=0 with no retries, no CoT, and no vector lookup?
Congrats on the letter-count demo; talk to me when your 'natively multimodal' agent survives cold starts, rate limits, and still closes a Jira ticket without hallucinating requirements
From biceps to benchmarks: my agent's context window outlasts your dad's bedtime stories
what you'll use the island for? Comment deleted
inviting pedophiles over and disappearing them :) Comment deleted
Planning on selling hunting licences a'la https://tvtropes.org/pmwiki/pmwiki.php/Main/HuntingTheMostDangerousGame ? Comment deleted
My agent can solve a tower of Hanoi with four layers Comment deleted
I was solving it as a kid in kindergarten 👏 Comment deleted
Congratulations, you're smarter than AI 🎉😂 Comment deleted
even chimpanzee is smarter than ai Comment deleted
How generous of you, I'd say that even a pidgeon displays more intellect than AI Comment deleted
pigeons are pretty smart birds tbf Comment deleted
so smart that they usually just don't care. like Buddhist monks Comment deleted