Jesse Alama reposted
Debates over whether specs or code are the ultimate source of truth in ai assisted software applications are reminding me of the Richard Cook bit about the system as imagined and the system as found https://youtu.be/2S0k12uZR14
Jesse Alama
@jessealama.net
Math and programming. Lean, JS, Racket. American. Igalian.
183 followers60 following174 posts
Jesse Alama reposted
Debates over whether specs or code are the ultimate source of truth in ai assisted software applications are reminding me of the Richard Cook bit about the system as imagined and the system as found https://youtu.be/2S0k12uZR14
Jesse Alama reposted
I love formal verification + fuzzing techniques but am wary that we're quickly speedrunning towards a future where people are generating enormous incomprehensible correctness proofs about their programs, but are none-the-wiser on what properties have actually been proven.
I don't say this lightly: this approach is clearly the future of software development. I have a blog post brewing that touches on the subject, but fuzz testing and formal verification at scale will make all software an order of magnitude better, safer, and more reliable.
Pabst is a fancy frontend to fast-check. The core property-based technology lies there. fast-check.dev
I was inspired to make pabst after organizing a session at @igalia's Web Engines Hackfest, dedicated to the general topic of getting better at testing our JavaScript. github.com/Igalia/weben...
Taking off now to A Coruña for the 2026 Web Engines Hackfest! Tons of great stuff on offer. Take a look at the program:
Watching Terence Tao's cleanup of a Lean proof to meet Mathlib guidelines, I'm struck by the problem we're in of there being too much math. There are so many good Lean proofs coming out but essentially only one blessed outlet for them. "MathOverflow" indeed! www.youtube.com/watch?v=l3SC...
I'm in Tsukuba, Japan for FLOPS 2026! After last week's Formal Methods symposium, this will be another packed week of learning. I'll be presenting a tutorial on Lean, extracted from my TypeScript-to-Lean compiler. functional-logic.org/events/flops...
Jesse Alama reposted
New blog post: On the Unreasonable Effectiveness of Property-Based Testing for Validating Formal Specifications. proofsandintuitions.net/2026/05/18/p... The gist: randomised testing can validate formal specs. It's very cheap and powerful: we found bugs in specs of VERINA and CLEVER benchmarks.
Getting ready to attend day 1 of AI, Proof and Verification (AIPV) at Formal Methods 2026 in Tokyo. aipvconf.org
I'm on my way to Japan to attend Formal Methods. The agenda is absolutely packed with great stuff; really excited to learn a ton. It'll be my first time at FM. If you're interested in Lean & program verification, let's talk! conf.researchr.org/home/fm-2026
I'm happy to announce Thales, a TypeScript compiler and JS engine in Lean. Thales compiles a subset of TypeScript to Lean via a shallow embedding. Thales is a bridge for TS programmers into Lean's program verification toolset. Check out github.com/jessealama/t... to get started.
I'm happy to announce Thales, a TypeScript compiler and JS engine in Lean. Thales compiles a subset of TypeScript to Lean via a shallow embedding. Thales is a bridge for TS programmers into Lean's program verification toolset. Check out github.com/jessealama/t... to get started.
Just learned about the Lean kernel arena from this post by Leo de Moura. With all the attention in the #lean space going to new features, massive growth of Mathlib, new initiatives like CSLib, this is something I'd never thought much about. leodemoura.github.io/blog/2026-3-...
Chris Bailey has a neat book focusing on the Lean kernel. ammkrn.github.io/type_checkin...
Jesse Alama reposted
There are many tools for unprivileged sandboxing on Linux. You should probably go use one of them. But I wrote shandbox to scratch my itch muxup.com/shandbox /home/$user/sandbox shows up as /home/sandbox within the shared sandbox, which otherwise can only access explicitly mapped files/dirs.
Jesse Alama reposted
The program for Leaning In! 2026 is settled. We've got 11 (!) wonderful talks lined up. This is going to be a wonderufl time for digging into Lean, Registration is open, but space is limited so if you're thinking of coming, please do register soon. See you in Berlin! leaning.in/2026/
Amazing work on an Erdős problem (whose original statement needed to be tweaked), demonstrating the power of AI-assisted proving in math. mathstodon.xyz/@tao/1158558...
Lean Together 2026 starts today! It'll be four days of juicy Lean stuff. It's a remote-only conference, so feel free to join. See you there! leanprover-community.github.io/lt2026/
Really alarming to see this story of "LLM psychosis".I first saw it in a video by Nate Jones. It looks like there's an attempt to solve the wildly difficult Navier-Stokes problem, backed up by a Lean formalization. youtu.be/AzOJ9QLgfIk
💥 Breaking news from my Inbox 💥 I saved 251 hours on email with @SaneBox in 2025 #UnwrapYourInbox See for yourself 👉 www.sanebox.com/unwrap/2pktu...
TIL there's a JSON Schema validator for #lean at github.com/CAIMEOX/json... It seems to be abandoned, though (no activity for a year).
There's a PR to essentially overhaul the whole thing:https://github.com/CAIMEOX/json-schema-lean/pull/2
Enjoying the video from the December 2025 Mathlib community call.
Jesse Alama reposted
We''ve landed a few more talks since the last update! * Sorrachai Yingchareonthawornchai: The CSLib initiative * Simon Sorg: Machine learning for Lean * Jannis Limperg on Lean metaprogramming for AI * Will Turner on the new ProofBench effort Come join us for a day of #lean! leaning.in
Incredible results with #lean and Gemini Deep Think by Tao in the Erdős Problem. It feels like a lot of doors in the formalization space, whether that's math or hardware/software verification, are opening quickly. mathstodon.xyz/@tao/1155914...
The 2026 edition of the ITP (Interactive Theorem Proving) conference back in October featured a #lean event. It was great to be there and connect with so many Lean folks. The videos are finally available: www.youtube.com/playlist?lis...
Jesse Alama reposted
Leaning In! 2026 is ON! If you like #lean and are in Berlin, Germany, Europe, or anywhere on Earth, you are invited. We've got a room at Spielfeld in Berlin. See you on March 12th! More information, including tickets and the CfP, here: leaning.in/2026/
Fascinating talk by Kevin Buzzard about where he sees math going, why it has a big problem (for some time now, even predating the internet), and why #lean might be part of the solution. hackaday.com/2025/10/08/w...
Really excited to see the #lean Mathlib Initiative take off. Here’s their new web page. Mathlib is such a huge, and growing, project; this help from Renaissance Philanthropy is very welcome.
TIL In Lean you can output the axioms that a proof depends on using a simple `#print axioms` command.
Jesse Alama reposted
Even if you didn't have the chance to listen to any of the amazing Igalia Chats - it's time for you to start with a really good chat about #Webassembly, with the one and only @wingolog.org . open.spotify.com/episode/0DmQ...
I was reminded of the Blueprint tool for sketching mathematical profs in Lean. I haven’t used it yet but I’m going to see if this can help me structure my work. Also curious to see how Blueprint applies in other scenarios such as software verification.
Some reflections by Terence Tao on different levels, of formalization and rigor in math. This was offered as a discussion topic in the afternoon session of the Lean workshop at ITP 2026 but we didn’t really get around to it.
This is a real eye-opening article where @ghuntley.com builds a compiler for a new language using LLVM and the amazing technique he calls Ralph Wiggum.
I haven’t yet dived in to git worktrees but as I continue to learn and reflect more on AI-assisted working, I realize this is pretty much the feature i need. This article made things click.
Jesse Alama reposted
No more juggling CommonJS and faux-ESM. @joyeecheung.bsky.social revealed how Node.js is moving to full ESM and why the future looks brighter for developers. #NordicJS #NordicJS2025
Jesse Alama reposted
OOPSLA 2026 deadline is just a week away! Can you please help us get the word out by sharing? Thanks! 2026.splashcon.org/track/oopsla...
Just arrived in Reykjavik for the #lean workshop as part of ITP 2025 leanprover-community.github.io/itp-2025-lea... I'll be presenting some of my recent work. Looking forwrard to (re)connecting with some fellow Leaners!
TIL pre-commit.com I was just browsing the #lean Mathlib repo toplevel directory and noticed a file with an unusual, but suggestive name. Took a look inside and...love it!
I'm happy to announce that I've finally wrapped up a proof of Euler's polyhedron formula ("V - E + F = 2") and submitted a PR to Mathlib. I've been working on this in bits and pieces on evenings and weekends. github.com/leanprover-c...
I was greatly assisted by Claude Code and LeanSerch (leansearch.net).
The Kimina theorem prover at Project Numina is an amazing piece of technology. This kind of targeted, special-purpose LLMs seems to be just what is needed. They don't even need to be large; "small" LLMs might be just fine. demo.projectnumina.ai