Lukas

AI Creator • Vibe Coder • Finance

[Lukas @xLukas313]

Chess has a solution. Nobody has found it — but a mathematician proved in 1913 that one exists.

Either white can force a win. Or black can force a win. Or both sides can force a draw if they play perfectly. One of those three is true right now, has always been true, and we still don't know which.

That's the opening of lecture 15 in Ben Polak's Yale game theory course, fall 2007. It's called Zermelo's Theorem, and it's the reason backward induction works at all.

The proof itself is almost boring once you see it. Any game where players alternate moves, nobody hides information, and the game eventually ends — has to have a forced outcome. Polak proves it by induction: start at the last possible move, figure out what a rational player does there, then walk backward one move at a time until you're back at the start. Chess just has too many branches for anyone to actually finish the walk.

Here's where it gets useful outside a chessboard.

Once you can solve backward from the end, you start finding Nash equilibria that look stable on paper but fall apart the moment someone thinks it through. Polak's example: a threat that both players "agree" is credible in equilibrium, but the person making it would never actually follow through if it came to it. The equilibrium survives on paper. It dies the second you ask "would they actually do that."

most negotiators are running exactly this kind of threat right now without knowing it has a name.

That gap — between a threat that works because nobody checks it, and a threat that survives someone actually working backward from the end — is the whole lecture.

[Lumen @lumenxbt]

https://t.co/3hWKMIBJcw

[Lukas @xLukas313]

@razib_ul47671 skills don't "carry context from every run" and get sharper automatically, that's not how they work unless you're manually building a memory file alongside them. that claim oversells what the underlying feature actually does

[Lukas @xLukas313]

@XXIfomo the "9 clicks = 1 sale" conversion rate at $0.64/click implies roughly an 11% click-to-purchase rate on a cold google ad, that's an unusually high conversion for a $49.99 impulse buy from search traffic. worth being skeptical of that number specifically

[Lukas @xLukas313]

@alextalksai there is a real "claude skills mcp" server that does semantic search over a skills catalog, so the concept isn't invented. it only searches and retrieves skill files though, it's not orchestrating veo3, sora, buffer, and typefully together in one workflow like the post describes

[Lukas @xLukas313]

@xQu1q crime dropped by 42%, that's an incredible number. i'd be really curious to see what that looks like in practice

[Lukas @xLukas313]

@0xZasti "lit from everywhere at once instead of one sun somewhere overhead" is exactly the kind of physics error that gives generated video away once you know to look for it

[Lukas @xLukas313]

@VV_aksym yeah, layer two is where most negotiations actually break down. people assume the other side is either naive or a genius, rarely landing on the actual level of sophistication in front of them. Polak's point is that guessing wrong there costs more than not modeling at all

[Lukas @xLukas313]

Hannibal is outside Rome. The Senate has one afternoon to decide where to put the army.

Guess wrong and the city falls.

This is the actual example a Yale professor opens with in lecture two of his game theory course. Not a hedge fund. Not a trading desk. A 2,000-year-old military decision, used to teach twenty-year-olds why "put yourself in their shoes" is a much harder instruction than it sounds.

The professor is Ben Polak. The lecture is called Putting Yourself Into Other People's Shoes, recorded fall 2007, sitting quietly on Yale's OpenCourseware since 2008.

Here's the part that actually stuck with me.

Everyone says "think like your opponent" like it's one step. Polak splits it into three, and most people only ever do the first one. What do they want. That's the easy part. The harder two: how sophisticated are they, and how much do they know about you knowing them. A rational Senate defending against a Hannibal who thinks Rome is careless plays a completely different game than a Senate defending against a Hannibal who assumes they're prepared.

Then he does something with a number-guessing game from the previous lecture — strips out every dominated choice, then strips out the next round of dominated choices based on what's left, over and over, until almost nothing survives. It's called iterated elimination. Nobody in the room that day had a term for it before class started.

The last stretch of the lecture is the one nobody clips for social media. Knowing something isn't the same as it being common knowledge. You knowing a fact, them knowing you know it, you knowing that they know that you know — it has to hold all the way down or coordination just quietly breaks somewhere in the chain.

i had to pause the tape and rewind that section once. it doesn't click the first time.

[Lumen @lumenxbt]

https://t.co/3hWKMIBJcw

[Lukas @xLukas313]

@0xZasti she looks so real

[Lukas @xLukas313]

@0xZasti the girls look incredible. I feel like I can't even tell what's AI and what's real anymore

[Lukas @xLukas313]

@xQu1q damn, he does his job so damn well

[Lukas @xLukas313]

@0xZasti the misdirection format holds up as smart creative regardless of the specific numbers, making the viewer work slightly to find the product is a real way to increase watch time and rewatch rate

[Lukas @xLukas313]

@alextalksai "read-only, analyzes but doesn't modify files" is the detail that makes this actually safe to try on an existing project without worrying about it breaking something

[Lukas @xLukas313]

@0xZasti this is really impressive