Somewhere, something incredible is waiting to be known -- Carl Sagan
Grenoble
Joined March 2008
- Tweets1.2K
- Following886
- Followers154
- Likes3K
Yannick L. retweeted
Modularity abstraction versus Modeling abstraction
muratbuffalo.blogspot.com/20…
Yannick L. retweeted
Yes, that was the fix that most DBMSes applied for fsyncgate
But most still don’t use Direct I/O when they replay the WAL on reboot.
So they risk externalizing non-durable data as durable (because they still read from the incoherent kernel page cache).
cf. “Can Applications Recover from fsync Failure?” which looked into this in 2020, two years after these fixes, and found that PG, SQLite et al were still vulnerable.
Yannick L. retweeted
Niklas (@dergoegge) outlines an approach and a solution, bedrock, for this problem of determinism.
He focuses on the challenge of emulating time and pausing a VM at an exact instruction count, something CPUs aren't designed to do.
brink.dev/blog/2026/06/25/be…
Yannick L. retweeted
Three ways to drive testing from a Quint spec, depending on how much control your runtime gives you:
→ Model-Based Testing: replay traces from the model
→ Hybrid: model proposes the next action based on results from runtime
→ Trace Validation: capture from production, check it against the model
A peek at @bugarela's BugBash by @AntithesisHQ 2026 talk. Full video soon!
Yannick L. retweeted
New 0-day in Lean just dropped. Your secure software just got borked. All your bases are mine.
kirancodes.me/posts/log-who-…
ALT Lean proved this program was correct; then I found a bug. I fuzzed a verified implementation of zlib and found a buffer overflow in the Lean runtime. AI agents are getting very good at finding vulnerabilities in large-scale software systems. Anthropic, was apparently so spooked by the vulnerability-discovery capabilities of Mythos, they decided not to release it as it was "too dangerous" (lol). Whether you believe the hype about these latest models or not, it seems undeniable that the writing is on the wall: The cost of discovering security bugs is collapsing, and the vast majority of software running today was never built to withstand that kind of scrutiny. We are facing a looming software crisis. In the face of this oncoming tsunami, recently there has been increasing interest in formal verification as a solution. If we state and prove properties about our code using a mechanical tool, can we build robust, secure and stable software that can overcome this oncoming barrage o
Yannick L. retweeted
i like to think of settings like a pressure valve
every time you add one its because you avoided making a real product decision
if youre not careful your product becomes a graveyard of avoided decisions
lets just add a setting often means we dont want to pick a default
it might feel diplomatic in the moment
everyone gets what they want in the end
except for the user
they end up with a minefield of toggles they dont understand
good products make strong choices
they say: this is how it works by default because we believe this is the best way
settings should exist for real variability
not to protect us from disagreements
when you look at an app with 200 settings youre not seeing power
youre seeing a decade of unresolved arguments
and its costing clarity for you and your customers
im not anti settings
im anti cowardice
if the only reason a toggle exists is: we didnt want to take a stance
it probably shouldnt exist
Yannick L. retweeted
Old KPIs protect old products.
A lot of strategy is just refusing to pay the migration cost.
"Specy addresses this challenge by automatically
learning specifications from observable event traces generated by message exchanges in distributed systems." 🤯. I'm a huge fan of P. I can't wait to play with this.
TL;DR: we identified a pattern of FOL formula over message events (exchanged between servers) that can express various safety properties of distributed systems, and we combine static&dynamic learning to effectively discover them.
Check out our pre-print: cs.princeton.edu/~dh7120/ass…
Yannick L. retweeted
Happy to announce that our paper “Specy: Learning Specifications for Distributed Systems from Event Traces” has been accepted at OOPSLA’26. This is a joint effort with Ankush (@ankushpd) and the P team! It’s not possible without all the great collaborators!
Yannick L. retweeted
"Scale isn't just performance or distribution, it's survivability." - @TigerBeetledev's @jorandirkgreef. At Monster Scale Summit, find out what keeps most transaction processing systems from scaling and how to survive billions of transactions. ow.ly/wPbj50Yoyjz
#ScyllaDB
It's so easy to decide on a name youtu.be/4hMXaxF9kVI?si=shW9…
Yannick L. retweeted
Stop worrying about slop
For a specification φ a coding agent must find a program p such that p meets φ
p ⊨ φ
Slop is what we call code that offends our taste. But style is a preference. Correctness is a requirement.
Correct slop is still correct
xitter.cf/DominikTornow/status/2…
Yannick L. retweeted
Most developers take for granted that software's going to have bugs. But why?
A new perspective to start the new year. This was Will's opening talk at #BugBash last year - join us on April 22-24 for this year's edition!
Full links in the comments.
Yannick L. retweeted
✨ Magie du soir sur la chaîne de Belledonne…
📸 Photo prise par myplaneterre sur Instagram ✨
Petit rappel : ces paysages sont précieux, prenons-en soin.
#monisere #ProfitePartageRespecte #Moucherotte #Belledonne #Isere #Grenoble #Montagne
Yannick L. retweeted
"I doubt that anything resembling genuine "artificial general intelligence" is within reach of current AI tools. However, I think a weaker, but still quite valuable, type of "artificial general cleverness" is becoming a reality in various ways."
~ Terence Tao
Even if the internal method is kind of janky, you can still get a high success rate if you pair it with strong verification.
The model throws out lots of candidate approaches, and then you use strict filters to reject the bad ones. When you do that at scale, you get something that can beat what a single human can do just by volume and selection pressure, even if each individual attempt is not very trustworthy.
That leads to the vibe people have right now: the tools feel impressive and useful, but also unsatisfying.
The output might still be technically valuable, but it stops feeling like “mind”.
Yannick L. retweeted
Property-based Testing and Deterministic Simulation Testing change how you think about invariants:
First you defend the invariants your system has. Then you design the invariants your system should have
Invariants become aspirational, not just defensive
Great session with @DominikTornow on deterministic simulation testing. @resonatehqio is building a truly unbreakable foundation for distributed systems. Highly recommend keeping an eye on what they're doing! 🛠️
Yannick L. retweeted
Each programming language I used taught me something:
1. C++ taught me how powerful abstractions can still be performant
2. Python taught how people are eager to give up performance to write glue code
3. C taught me that simple but solid foundations can get you really far
4. Java taught me that enterprise software is unstoppable
5. Haskell taught me the full power of Functional Programming
6. OCaml made me realise that having that power is dangerous, and you want a pragmatic subset.
7. CoffeeScript showed me that making a language that's just a syntax sugar for another language only makes things worse
8. Go
9. JavaScript taught me that it's a bad idea to build the foundations of web development on top of a language designed in 10 days
10. TypeScript showed that adding static typing to a dynamic language is a noble goal, but it's too late.
Exactly
i am convinced that software devs have a speed problem
they think the #1 issues is writing code faster... its not. its fixing the code that is already there to stop being utter garbage (as a garbage code connoisseur)
quality is really lacking these days, yet quantity has never been higher