William Meng

Working on AI in AEC (YC-backed)🏠👨‍💻 | Tinkerer 💡 | Armchair observer of political trends & philosophy 👀

[William Meng @wm__meng]

@boson2photon Lol first 2 are easy

Got me on the 3rd one tho… would getting into YC count even more against me than merely applying 😂

[William Meng @wm__meng]

@quaboud @sophieisawitch If it’s possible to RL on solving a math problem, it’s possible to RL on finding mistakes in a solution

They are flip sides of the same synthetic data problem

[William Meng @wm__meng]

@longmarcher @sophieisawitch If Newtonian physics was provable, then relativity would never exist

[William Meng @wm__meng]

@quaboud @sophieisawitch The best one can hope for may be another adversarial verifier agent whose reward is to find mistakes

The problem of technically unsophisticated political patrons trying to make decisions based on experts he cannot fully trust is a problem as old as time

[William Meng @wm__meng]

math research could ultimately resemble industrial bounties, where engineering orgs contribute bounties on problems it cares about, and math orgs are financed to solve them

[William Meng @wm__meng]

@KindlyContrary @AndrewCurran_ All medical treatment have the risk of unforeseen side effects

Unlike math, most of the empirical sciences have no provable truths, only highly likely guesses

Math will just have to learn to tolerate the same ambiguity

[William Meng @wm__meng]

The “scientific community” is starting to sound like the catholic monastery desperately holding on to relevance in the face of the printing press and reformation

If AI produces more useful results than academic, people will follow the AI for downstream application

There is no final arbiter of scientific truth. People can believe whatever they want to believe.

What are you going to do? Burn the heretics?

[Majid Hosseini @m4h007]

In my view, the right response by the mathematical community to OpenAI’s papers is to say:

“We’re not accepting these papers as finished work. It’s not our job to sit down and read your model-generated proofs.

It is your job to present your work through presentations and convince people of the correctness of your work. It’s always the case that whoever makes a claim needs to provide sufficient explanation and evidence to make independent verification feasible.

Until that moment, you just have a claim.”

There’s a difference between a proof being mathematically correct and a proof being accepted as correct by the mathematical community. Mathematical truth doesn't depend on consensus, but scientific recognition requires scrutiny, communication, and reproducibility.

They're still the arbiters of whether anyone's claims, including OpenAI's, are accepted as valid, and they should leverage that to the maximum extent possible.

[William Meng @wm__meng]

Empathy is a tool, and it doesn’t imply compassion. This sort of casual mixing of empathy and sympathy is giving empathy a bad rep and creating strategic blindness by having no theory of mind for one’s enemies, because people are uncomfortable with “empathizing with the enemy”

You can empathize with your enemies. And in fact, it’s often a strategic imperative to empathize with one’s enemies, to know your enemy the way they know themselves, so that you can learn to destroy them

https://t.co/wFUEqp7bEo

[William Meng @wm__meng]

@realBathshiba @DrewPavlou Please distinguish the difference between sympathy and empathy first

Empathy is a form of knowledge, it has nothing to do with morality

[William Meng @wm__meng]

We live in an over sanitized society that has created a generation of cowards the likes of which history has never seen

The great issues of the day are ultimately decided by iron and blood (or the fear of it)

Our laws and political order would have no meaning if they’re not backed by collective willingness to impose coercion

It’s healthy to peel back the veil and let the public see how the sausage is made once in a while

[William Meng @wm__meng]

@CallMeEzekielYT You’re cherry picking data points here. The Chinese Civil Waf would prove the inverse

[William Meng @wm__meng]

AI can prove a lot of lemmas along the way. If we globally index those lemmas, even if the problem in the end wasn't solved, other AI agents can build upon those lemmas to do new things

In other words, reinventing academic publishing from first principles, by agents, for agents, and peer reviewed by agents. Humans are merely the board trustees

[Race Horse @punishhorse]

@arctotherium42 I have heard the argument that a lot of progress and useful innovations in mathematics come as a byproduct of trying to solve problems, which is lost when you just tell AI to do it.

After all, where do the next "maths problems we need to solve" come from?

[William Meng @wm__meng]

How does a grade schooler reliably know if a mathematician’s proof of the Riemann hypothesis is correct?

We are all going to be grade schoolers compared to AGI

At a certain point, the amount of math problems that demand to be solved will exceed the total human cognitive capacity of our human capital, but that doesn’t mean we shouldn’t pursue the collective capacity of our civilization further.

Just like how our industrial demands have far exceeded the collective capacity of all human muscle

[William Meng @wm__meng]

How does a grade schooler reliably know if a mathematician’s proof of the Riemann hypothesis is correct?

We are all going to be grade schoolers compared to AGI

At a certain point, the amount of math problems that demand to be solved will exceed the total human cognitive capacity of our human capital, but that doesn’t mean we shouldn’t pursue the collective capacity of our civilization further.

Just like how our industrial demands have far exceeded the collective capacity of all human muscle

[ringleader @mljumaru]

@BunzaiM @provisionalidea Except if GPT or Claude prove the Riemann Hypothesis, you have no way of reliably confirming such a proof is correct. The whole point of science being falsifiable breaks down if no one capable of doing so is involved, no matter how many lines of Lean an LLM can hallucinate.

[William Meng @wm__meng]

Academia is behaving awfully like the church priesthood on the eve of the scientific revolution, desperate to hold on to their monopoly on knowledge and the prestige associated with it

[James Rosen-Birch ⚖️🕊️ @provisionalidea]

the mathematicians do not pull punches

[William Meng @wm__meng]

The funny thing is that authorities in China are almost the inverse, where they love the striver and despise the gifted slacker

My grandfather back in the 1950s didn’t do his homework and often skipped class cuz he thought it was boring, got a perfect score on gaokao math, which was like 1 in a million, and instead of going to top schools, his teachers despised him as a heretic so much that they recommended sending him to work in a frontier lumber yard

The authorities were afraid that the success of the gifted slacker would cause strivers to emulate the slacking off without being able to compensate with natural talent, thus breaking the fantasy of strivers overcoming a lack of natural talent

[William Meng @wm__meng]

@CheyenneHuntCA If there’s a drug that can help middle-aged men stay physically attracted to their aging wife instead of feeling the urge to leave and date some 20 year olds, would you characterize it in the same way?

[William Meng @wm__meng]

My point is that the mavericks/trailblazers inherently have nothing to fear from strivers, since mavericks go after contrarian bets instead of consensus bets

Most conformist strivers go after the same narrow niche. If you go after the same thing as strivers and lose, then you’re by definition not a maverick, just a less competent striver. And if you do something contrarian, most strivers won’t compete with you there anyway

And since contrarian bets are high risk high return, the apex elites will always be the mavericks for as long as we are a capitalist country