Kaworu

Alpha? Here? No
Just things about AI

[Kaworu @kaworu0x]

@0xkair0s this is why i never trust a simulation on its own

[Kaworu @kaworu0x]

@crimson_moti the gym clip is genuinely hard to tell

[Kaworu @kaworu0x]

@LyckyOus local for the boring stuff makes a lot of sense

[Kaworu @kaworu0x]

@Yogabool i’m more curious how long people stay once they find out

[Kaworu @kaworu0x]

@crptAtlas construction worker was the missing niche apparently

[Kaworu @kaworu0x]

@fuchexcrypto the 3am replies are the creepiest part

[Kaworu @kaworu0x]

The mathematician who taught computers to correct their own mistakes spent the rest of his career studying one they could never correct: solving the wrong problem perfectly.

Richard Hamming worked on the Manhattan Project, shared an office with Claude Shannon at Bell Labs, invented error-correcting codes, and won the Turing Award. In 1995, three years before his death, he gave one final lecture to a room of students.

It was not about code.

It was about why equally intelligent people can work equally hard for forty years and produce completely different lives.

Los Alamos, 1945.

Hamming ran the calculating machines while Feynman, Fermi, Teller, Oppenheimer, and Hans Bethe made the decisions. He called himself the janitor of science: trusted to produce answers, never trusted to decide which questions mattered.

He became obsessed with one problem. These men were not obviously smarter than everyone else. They did not work longer than everyone else. So why did their work change history while other brilliant people disappeared?

Bell Labs gave him the answer.

Hamming began joining different groups at lunch and asking what they believed were the most important problems in their field. Then he asked which of those problems they were actually working on.

Most had no answer.

Finally, he asked the question that got him thrown out of one lunch table:

“If what you are doing is not important, and if you don't think it is going to lead to something important, why are you working on it?”

One scientist listened. He spent the summer thinking about the question, later became head of his department, and eventually entered the National Academy of Engineering. Hamming said he never heard anything significant about the rest of that table.

Intelligence compounds only after it is pointed at something worth compounding. A perfect solution to an irrelevant problem remains irrelevant. More effort does not rescue bad direction. It takes you farther in the wrong direction.

Hamming had seen the same failure inside machines. In the 1940s, he received access to a Bell Labs computer on weekends. If one relay failed, the machine detected the error, abandoned the calculation, and moved on. Hamming would return Monday to discover that an entire weekend had produced nothing.

So he invented a code that allowed the machine to locate and correct its own errors. That idea now protects computer memory, data transmission, and the digital infrastructure beneath modern AI.

Hamming taught the machine how to recover when one bit went wrong.

He never found a code for recovering when the objective was wrong.

That is the part every AI system inherits from us. A model can search more possibilities, test more parameters, and optimize faster than any human team. It cannot prove that the target it was given deserves to be optimized.

Wall Street calls this sophistication: better models, more compute, more autonomous agents searching the same historical data for the same shrinking edge.

Hamming would have asked the question before opening the terminal:

Is this an important problem, or merely a problem the machine knows how to solve?

The lecture is free. The code runs inside the modern world. Most people will use both without noticing the warning.

AI can correct the bit.
You still have to correct the objective.

[Misato @misat0x]

https://t.co/55YhnZ76LI

[Kaworu @kaworu0x]

@elonmusk ahahahah really

[Kaworu @kaworu0x]

@misat0x wow,i'll quote it

[Kaworu @kaworu0x]

@misat0x really important problem

[Kaworu @kaworu0x]

@crimson_moti i hate how normal this sounds

[Kaworu @kaworu0x]

@misat0x

[Kaworu @kaworu0x]

An IBM mathematician discovered that the number Wall Street uses to measure risk might be infinite. It began with a cotton-price chart on a Harvard blackboard in 1961. The economist who drew it had failed to model the data and challenged him to take over. Benoît Mandelbrot solved the problem in weeks, then spent decades watching finance use the model he had broken.

The standard model assumed markets moved like a random walk, prices reflected available information, and daily changes followed a bell curve. Mandelbrot kept the first two and replaced the third. A market could be unpredictable without being safely Gaussian.

The cotton data showed why. Small changes happened constantly, but extreme moves appeared far more often than the bell curve allowed. Violent periods clustered together, followed by deceptive calm, then another burst large enough to make the previous risk estimate useless.

Mandelbrot’s replacement model allowed the variance of price changes to be infinite. That does not mean every asset carries infinite risk. It means the standard deviation used to measure that risk may never settle into the stable number the model assumes. One extreme observation can still destroy the answer.

In 1963, he published “The Variation of Certain Speculative Prices.” The paper began with cotton but applied to every market that treated a crash as an error outside the system. By 1982, Mandelbrot wrote that economists were citing his work to acknowledge the limitations of Gaussian models, then continuing to use them anyway, possibly “to assuage his conscience.”

He left economics and created fractal geometry: the mathematics of coastlines, clouds, mountains, and turbulence. These structures keep their roughness at every scale. Prices did the same thing. Remove the time label from a market chart and it becomes difficult to tell whether the movement occurred across seconds, days, or years.

MIT, 2001

Mandelbrot puts real price series beside charts generated by his multifractal model and calls the synthetic one “my current forgery.” Most people cannot identify it. The fake contains the same quiet stretches, volatility clusters, and enormous spikes as the real market because it models what the bell curve deletes: extreme movement is part of the architecture.

At the end of the lecture, someone asks whether fractals can describe human behavior. Mandelbrot answers:

“Certainly, stock market is human behavior.”

Fear clusters. Leverage clusters. Forced selling clusters. Markets do not produce clean independent observations because the people inside them are reacting to one another.

[veles @velesxbt]

https://t.co/oHZTyjGo7u

[Kaworu @kaworu0x]

@Nekt_0 CI repair is the one i’d automate first

[Kaworu @kaworu0x]

@Yogabool the reveal would be brutal😂

[Kaworu @kaworu0x]

@misat0x it's rly cool

[Kaworu @kaworu0x]

@misat0x 80 20 hm it's interesting part

[Kaworu @kaworu0x]

@Yogabool the fake girlfriend economy is getting very real

[Kaworu @kaworu0x]

@misat0x she looks real and cute(

[Kaworu @kaworu0x]

@rftd09 i’ll believe the iPhone line after mainnet