AI Conferences: A Costume Party for People Who Can't Spell Gradient

AI conferences seem to be nothing but costume parties where speakers dress up as data scientists but couldn't tell you what a gradient is. They'll say tensor thinking its a setting on a blender and are responding to their lunch order.

The speaker lineup is people whose entire AI experience seems to be wrapping API calls in try-catch blocks and writing prompts that say "You are an expert." Now look, if your conference is teaching about try-catch blocks and how to type on a QWERTY keyboard, hell yeah - I've never been to a Typist Conference but now I'm interested.

Want to understand transformers? Don't ask because you're asking someone who's never computed a softmax. Want to know why your model hallucinates? You're getting answers from people who think temperature is just a slider that makes things spicier, so bring your own tobasco. Probability and statistics? Wrong conference! How to press enter efficiently? They've got answers.

Remember that automotive engineering summit where the speakers were Uber drivers? I don't. But I bet it went something like this: "Let me tell you about the engine." Great, what about it? "You push the gas pedal."

The actual ML engineers, the statisticians, the people who understand why attention mechanisms work and when they don't? They're not on stage. They're too busy doing the work. Meanwhile the keynote slot goes to someone whose deepest technical contribution is setting max_tokens to 4096.

Before I forget, whens the first Crayon Eating conference? I'm predicting 2027.

Back to the Typist Conference. Do you give out WPM awards? You should live demo Mavis Beacon Teaches Typing because it was so much fun and was also something I had to do daily if I wanted to goto my friends house and play SNES. I miss 1987.

WPM: 120

❤️
Jake

Song