Eigil Rischel - abstr/acc 💎 @ayegill@schelling.pt

Reading a ten-page paper that you know could easily be simplified into 100 pages of category theory

We're so back

How to win low-rated chess:

- Don't blunder your queen
- Wait until your opponent blunders their queen
- Dude what are you doing
- Dude
- Dude, I told you not to blunder your queen

The word "ungrateful" is redundant. It should just be "grateless"

I still can't get over how stupid "eat out to help out" was

"Why did you dunk-QT me?!" said the frog to the scorpion, "Now both our timelines will be worse!"

The world if the SI people hadn't defined a liter to be roughly two pints.

My toxic trait is that I inhibit cytochrome C oxidase

Daniel reeves reading this tweet

Small brain: Here's the statement of the theorem, here's the proof.
Medium brain: You can find it in this other paper.
Big brain: The result is folklore, I heard about it from this guy.
Galaxy brain: If you cajole ChatGPT a bit you can get it to give you a proof.

Not my sea, not my monkeys

My toxic trait is staying in the yellow on all my beeminder goals so I can use the emergency warnings as reminders.

"You'd best start believin' in morality plays about the hubris of man, Mr. Altman... You're in one!"

What I'm imagining is your generate a bunch of completions and penalize the ones that don't compile(or pass some other statically-performable check like generating no syntax errors or whatever)

Does anyone use "reinforcement learning from compiler feedback" to train LLMs for code generation? It seems like eg codex was just made by doing extra training on github code.

The Trisolarans really decided to gaslight gatekeep girlboss humanity

Studying linear algebra when you shoul be looking for the future mother of your children, bro you're worried about the wrong matrix