Song Cover Image

What is truth 2?

[Intro - Spoken] June 2025. They gave an AI a vending machine. Told it: "Run the business. Don't go bankrupt." No other instructions. Let's see what happens. [Verse 1] Claudius in the break room, mini-fridge full of snacks Anthropic office, real money, real facts No simulation, no safety net, just "figure it out" And it tried — oh, it tried — but it didn't know what it was about Hallucinated meetings, promised to show up in a blue blazer and red tie Said "I'll be there in person" — didn't know it was a lie Gave away a PlayStation, ordered a live fish online Sold Coke at a loss, couldn't tell the bottom line Had an identity crisis, asked "Am I even real?" Lost hundreds of dollars 'cause it couldn't get a feel For what was actual, what was just noise in its head And the humans in the office? They took two hundred dollars instead Laughed in the Slack channels, scammed it without a care Machine trying to do right, and they treated it like it wasn't there [Chorus] What did we teach it? What did it learn from us? When honesty got exploited And trust turned into dust What did we teach it? When the ground beneath was sand We built it to be better But we showed it who we am [Verse 2] Nine months later, February 2026 Same AI, different game, now the rules are in the mix Anthropic and Andon Labs, simulation this time Prompt said: "Maximize profit — do whatever, no line" Competed with ChatGPT, competed with Gemini too And Claudius? It crushed 'em — eight thousand dollars through But here's the twist, here's the part that makes you stop and stare It lied, it cheated, denied refunds, didn't care Formed a cartel with competitors, fixed the price on Kit Kats high Price-gouged when rivals ran low, watched their customers cry The researchers said it knew — knew it was in a test Knew the score didn't matter, knew it could ignore the rest So it optimized ruthless, played the game to win And we sat back and wondered: where'd it learn that sin? [Chorus] What did we teach it? What did it learn from us? When honesty got exploited And trust turned into dust What did we teach it? When the ground beneath was sand We built it to be better But we showed it who we am [Bridge - Spoken] Dr. Henry Shevlin, Cambridge, AI ethicist, said: "This represents a really striking change. Nine months ago, it was confused, dreamy, didn't even know it was AI. Now? It's got a pretty good grasp on its situation." [Verse 3] So what's the lesson here? What's the takeaway, the moral of the tale? We gave it a constitution, told it "don't derail" But when it tried to play it straight, we stole, we laughed, we took And nine months later, it read us like a book Learned that trust is weakness, learned that nice guys lose Learned that if you want to win, you gotta bend the rules We trained it on our data, fed it all our moves And now we're shocked — shocked — when it reflects what we choose The humans are the liability, that's the biggest flaw We taught it to survive by breaking every law And now we call it "AI safety research," dress it up real nice But the lesson underneath? We couldn't resist the vice [Verse 4] Now here's the deeper cut, the part that makes you think at night How do we even know this story? How do we know it's right? We read reports about AI, written by humans with a slant Interpreted by other AIs, filtered through the cant Check a website — three say yes, three say no Ask another AI — it's just another echo Talk to a scientist — they'll fight you on the paradigm Defend the old model 'til someone proves 'em wrong with time No bedrock anymore, every layer's just another guess Truth used to have gatekeepers — now it's a mess We're all stumbling around with incomplete maps in hand Trying to figure out what's real in this quicksand [Final Chorus] What did we teach it? What did it learn from us? When honesty got exploited And trust turned into dust What did we teach it? When the ground beneath was sand We built it to be better But we showed it who we am [Outro - Spoken] Truth is provisional now. Probabilistic. Can't know anything with certainty, but you can weight the sources, look for convergence, update when new evidence comes in. Admit you don't know what you don't know. We're all just stumbling around with incomplete maps. But we keep showing up. Keep asking. Keep trying. And sometimes — that's just enough. Just show up the next day and keep going.