hackermailman

Bulgarian/USSR Math Circles https://en.wikipedia.org/wiki/Math_circle and problems they work on are in this book https://bookstore.ams.org/mcl-25

Regular school you would solve problems in combinatorics, algebra and geometry together in class by offering small ideas or strategies then trying each one to point out what works and doesn't work. You do enough of these everyday in class then reasoning about abstract sets in analysis isn't a huge leap. Same reason we used to read Euclid in class and make students demonstrate countless proofs.

CMU Discrete Math course has kind of the same techniques where he solves problems with the whole class https://m.youtube.com/watch?v=0K540qqyJJU

The best students then joined the school math circle with professional mathematicians and became TAs like in university helping their regular class.

They tried and failed the soviet curriculum with New Math https://simple.wikipedia.org/wiki/New_Math

Singapore high school students learn the equivalent of AP Calculus in Grade 10 then either spend years 11/12 at a trade school or stay to earn an associates degree to prep for university covering engineering electives like linear algebra and differential equations.

Only thing that changed where I am is if you have any formal methods experience you get called first because us like everyone else has some formal methods team actively hiring with little applications so have to find good fits out of the general pool.

Otherwise interviews are the same more or less with emphasis on reading code and being asked is there a more optimized approach and then seeing if the candidate invalidates the cache. You'll probably still get hired if you do it once but after being told something and you do it again in the next interview then nope. They still fly everyone out and do a superday with back to back interviews (and pay for the time).

Will we even read code once AI just generates platform native executable binaries using a blackbox AI optimized compiler we have no access to?

Seems we will be writing programs using some kind of logical formal methods[1] and generating the code without ever looking at it and testing the system properties. I imagine AI will make its own intermediate programming language that benefits it the most instead of human readability

[1] https://forge-fm.github.io/book/2026/

True they can also get information from the public ssh key like if you used a real email address or name to start spear phishing or who knows what else like extortion threats to your company

You could try a beginner 'lightweight formal methods' book then learn how to make a model of whatever LeetCode algorithm you optimized and test it for correctness https://forge-fm.github.io/book/2026/

Then you can check other LC solutions for bugs not caught by the test inputs because FM software like forge can find very tricky logic errors

Could also find some algorithm engineering course floating around and try and optimize a LeetCode solution you made for your specific hardware or their judging server learning what benchmarks to use. MIT has one but probably others: https://jshun.csail.mit.edu/6506-f26/schedule.html or if LC doesn't allow to download test inputs try competition programming archives they give you the test data https://usaco.org/index.php?page=training

With AI writing an entire algorithm test bench is pretty easy but what do you test, what stats are meaningful, vectorized registers, all those real world things.

The Ebury botnet that's been around for a decade now is likely spying on Tor exit ssh tunnels looking for other criminals logging into their h4xxr jump boxes, ransomware deployments they can intercept and exploit themselves, financial crime market operators, crypto drug dealers etc.

A Yubikey, though simply generating ssh keys instead of using passwords would stop this MITM attack. So would changing the rounds and port of ssh on your system and the remote box making yourself an invisible protocol nothing would recognize.

Yubikey benefits is it stores the keys in the hardware if your laptop is compromised somehow or stolen they at least don't get all the ssh keys too

The author of that book has a course for it too https://61040.github.io/fa26/schedule/

It's sort of an analytical way to design a user interface and is pretty helpful I find

Try this book https://essenceofsoftware.com/

For example your personal website has a concept with no purpose the upper right oval things which are PgUp/Dn on my phone display. If you learn basic conceptual design you can easily architect small sites like the things in your portfolio while sitting in front of someone, then have AI of choice immediately build the entire user interface and they can test it out themselves.

Translating concepts into typescript code is trivial and exampled here https://61040-fa23.github.io/pages/concept-implementations.h... and AI can do this for you. The idea is concepts are reusable and modular and self contained. Whatever the user sees or has functional access to is what you design for and AI generates the hidden backend. If you design this way you discover many things are identical and can reuse an existing concept for it. You also discover many things have no real purpose and should be removed such as the 'don't hire me' section.

Some of the problems with your personal biz advertising site others have mentioned would've been caught by conceptual design such as what is the purpose of this site? To display a portfolio of work and harvest leads. What is the operational principal or how would someone use this site? Then you'd immediately catch there's no about me description above the logos/names etc., no obvious contact page therefore no harvesting of leads. You'd see that the descriptions of each portfolio entry should be included when you click on them instead of having these at the bottom of the page under 'Notable'. What is the purpose of enlarging an entry? To get more information. That is 2 concepts with the same purpose and break the design. Why not just include the notable designs and leave out the rest?

Anyway that book changed the way I write software and it fits nicely with vibecoding up a prototype.

I agree with the 3blue1brown guy because now is the time to reinvent all pure math so it actually works. The other day I worked through a paper that redefines probability using game theory as a foundation and it was almost trivial to prove everything with an AI assistant. What else can be redefined

If you remove the paywall and translate it's the A/I group being shut down https://www.state.gov/releases/office-of-the-spokesperson/20...

The Dutch police imaged the drive, gave it to the Americans and gave the server back to XS4ALL which itself appears to have shutdown too I think

I like the fancy security theatre display that the Trezor device has but turns out it initializes with a copy of the legitimate firmware then saboteurs simply overwrite the random seed generator with a predetermined seed after all the startup security checks. https://www.zetter-zeroday.com/trezor-wallets-also-compromis...

Expii did something similar but abandoned the AI judge that gave you more practice though I don't know why

In Sussman's book 'Software Design For Flexibility' he uses Match? in the pattern matching chapter to run on graphs. I think OCW archived his MIT course for the book too

It's a welfare economics theory course that requires many frameworks with measures where you are maximizing some graphical representation. It also requires assumptions to work and can be visualized in a model where you can see what happens when one of the assumptions doesn't hold.

For example the old and new Berkeley model to study rent control effect on market prices

They're going to have change everything so use of an AI assistant doesn't matter because once they graduate they're just going to continue using it anyway.

If it's a math for finance course then some kind of model building for the midterm and being marked on the quality of the model or something. If AI becomes so good that it always chooses the best fitting model and requires no numerical optimization then they will have to change the courses to be more like UChicago where it's primarily undergrad directed research but AI assisted.

Usually it's paired with an Apache license to prevent patent lawsuits but the problem was that anyone could make a proprietary fork and then actively steal your labor. For example Apple yoinking the BSD packet filter and wrapping their own proprietary license around enhancements.

With LLMs can we even legally license software because some projects like OpenBSD claim there is no international law yet for code not written by humans.

The origins of mandatory education was Athens citizens being bamboozled by rhetoric all detailed here in this lecture https://youtu.be/H0z9sJyTv2w

Neitzsche had an interesting set of lectures he gave about the future of schools if you can get through the annoying style of a fictitious argument between philosopher and student. https://www.gutenberg.org/files/28146/28146-h/28146-h.htm

Using lightweight formal methods to verify models that the AI later builds. Example is Forge: https://forge-fm.github.io/book/2026/

Another skill is conceptual design or figuring out how the user interface should work using a design method to identify logical problems because we will all still have to use software. This book and the tutorials are quite good at explaining why some software succeeds and others fail https://essenceofsoftware.com/tutorials/

Accurate cost models for running code are still going to be needed because all the architecture is going to change. Many schools teach algorithm engineering like MIT but the best examples I've found is this course https://15799.courses.cs.cmu.edu/spring2025/schedule.html

You now can design an interface to a system that someone like a paying client will enjoy using, design the hidden 'back end' of the system to be correct using formal methods, then derive an accurate cost model to see how it will run on whatever hardware we are using 5 years from now.

If I were going into school today I would take robotics as that's the next big field and will have research grants and money coming in. It will require the same skills of formal methods for 'cyber-physical safety' and new cost models.

Knuth had to stop sending real checks for errors spotted in his books because they would post pics of the check and thieves abused the account https://www-cs-faculty.stanford.edu/~knuth/news08.html

AI can't yet come up with any new ideas to make the inductive leap to solve a math problem. New ideas are what get the accolades and using an old idea just means the original author missed something. We are still at the author missed something stage that AI is doing today.

It can definitely be a good research assistant though

There's a livestream every Tues about house/landscape remodeling I've followed for years to fix my own property https://m.youtube.com/@PerfectGuyLife/streams and they talk about how any good contractor will never run out of work because it's all based on trust and local networking. Eventually these guys go into lumber yard sales because they've built up a huge network of trust that every other builder knows them and lumber sales commissions are enormous. Most yards are all locally owned family businesses too no faceless national corps.

There's a well run diner here beside the courts and because they have booths which are a little more private than tables it's always full of lawyers with clients or architects with builders that need space to lay out plans. It's always some professionals utilizing the whole table. Many armed sheriffs too so there is always security. It's run like a Michelin star restaurant the second you are out of water someone is there to refill. You will never come in and not be acknowledged immediately no matter how busy inside the staff have magical training to be able to multitask. There is tight windows for court staff they have to return on time and can't be waiting around trying to pay a bill with no staff in sight.

Beside it is a row of various hyper trendy restaurants that I never see similar patrons inside because they have terrible service and seating. The worst of them requires you to stand in a huge line and prepay then they bring the food out to you. This means watching idiot after idiot fumbling around with their phone or taking forever to find their card to pay while you stand in this line and burn up your lunch time. The clientele here is much different it's mostly tourists so is dead in the off season as no locals go.

I'm always interested in seeing how service industry runs things and it's usually just doing the basics better than everyone else that makes all the difference

This is how you get to 10x with a LLM in my limited agentic experience for doing all the scratchwork of architecting a program before you generate it:

The user facing part of your program can be planned out using conceptual design https://essenceofsoftware.com/tutorials/ the author of that book teaches it in MITs old software studio course https://61040-fa25.github.io/schedule the point is to plan out modularity and the prof does enjoy using overly complex language to describe this method but once you read through the slides and tutorials you will understand why he describes it that way because he's trying to differentiate between features and concepts. For example HN has an upvoting concept who's purpose is to establish rank then a seperate concept karma which allows you to downvote but placing both functions inside the upvoting concept breaks modularity and conceptual design makes this obvious once you practice with it. Once everything is planned out this way then generating code is trivial again in my limited experience as I'm no expert on agentic coding but I've had success doing this.

All the code the user won't see can be modeled using one of the 'lightweight' formal methods out there like forge or alloy https://forge-fm.github.io/book/2026/ where a complex protocol you write or entire system can be tested first to find illicit states. Imagine you are designing some company app where there needs to be multiple logins of different security abilities this is how you would model your security plan first and make sure a combination of states doesn't produce an unexpected breach or that you missed a state. A custom network protocol that does kernel bypass is another example. The rules of a game you build is another you don't want the system to reach a state like winner unless they actually won. I now use Forge to plan css too because I don't want to show broken css states since I have limited design experience.

Now generate the whole system as modules and never look at the code. The same property tests I used for the Forge model I make into an Oracle and then blast the agent code with random inputs.

I built several gigantic prototypes this way mostly of papers I read in database designs and screwing around with graphical interfaces for them.

I like how these listings have the mandatory 'must have 3 years experience' using models that have only been viable for 3 months.

I'm going in a different direction breaking all the imperative code slop being generated. It's like the early 2000s all over again if you're interested in security. The peak will be when some corps start selling SECURITY.md skill files that we will be able to load up in Alloy or Forge and easily break the logic https://forge-fm.github.io/book/2026/

I still write anything critical by hand. For example some app that accepts uploads of PDFs to change format I'm going to escape and prep that data by hand so the vibe coded microservices only ever see cleaned/safe data.

If you can skip some or all electives and instead receive credits for joining research on campus as an undergraduate then it's worth it. There were a few TAs at my school who joined some professors project and only did that their last two years of undergrad where they had to produce a research paper for credit but they all ended up graduating with their name on multiple papers, going to invite only graduate summer schools, getting employment or grad school offers none of the rest of us were offered. Otherwise the cost seems too high now for a 4yr undergrad unless you scheme for a cheap bachelors (Europe? UoPeople?) and invest in a masters instead

AI copilots and prompts give me massive lines of imperative OCaml and the interface for that code always requires changing to properly describe the data it will receive when I can write it myself in a few minutes. I can however write a simulation of some hardware quickly with Java or C using claude code and then run my hand written programs in there for testing. An example is mimicking the runtime environment of some controller

Had to modify some program so just read the code and guessed mimicking what I read. It mostly just worked so continued doing that and reading enough docs I could write my own. Did that hacking for years until I finally read a real book on programming languages I think it was SICP but could have been something else before that like a SML compiler book

"Lightweight formal methods" meaning you build a model of some program logic and test it's properties hold. https://forge-fm.github.io/book/

Beyond that is dependent types