
TLA+
System softwareWikipedia
23
MENTIONS
4
EPISODES
4
PODCASTS
Search complete. 23 mentions across 4 episodes found for "TLA+".
Sep 8, 2026
Scaling your money safely with AI
R
3:13Ryan DonovanHOST
LLMs are not graded numbers, right? Yes.
R
3:15Ryan DonovanHOST
Something I've heard recently is using proving languages, like Lean or TLA+.
R
3:22Ryan DonovanHOST
Do you do any sort of validation through that?
S
3:25Srini VenkatesanGUEST
Not right now.
Mistral's Record €3B Raise Bets on Sovereign AI — Sep 8
J
2:13JamieHOST
They're a timeline problem for infrastructure teams tracking how fast they patch.
J
2:17JamieHOST
In the harness, tools, and orchestration world, Dan Lu ran coding agents through twenty-six different prompt conditions, roughly eighty runs each, all building the same Zstd compressor under instructions ranging from a bare default to test-driven development, property-based testing, and formal methods tooling like TLA+ and Lean4.
J
2:37JamieHOST
The default instructions beat most of the supposedly rigorous conditions outright.
J
2:41JamieHOST
Test-driven development actually produced worse tests that missed edge cases the default approach caught, and of eighty runs told to use TLA+, only five agents actually wrote the formal spec before writing code.
J
2:54JamieHOST
The failure mode is mimicry.
J
2:56JamieHOST
Agents copy a technique's syntax without performing its substance.
Replication Before Raft: 15 Years of Correctness at MongoDB
D
0:25David WynnHOST
Jesse talks about MongoDB's journey with formal verification, as well as his own.
D
0:30David WynnHOST
He shares what it was like to build a custom replication protocol using TLA+, and how he realized that you can work in TLA+, without a PhD.
D
0:39David WynnHOST
Jesse and Carl also talk about the future of formal methods, the impact of AI, the advent of simple but powerful tools like Antithesis, and Jesse's predictions about the future of code review.
D
0:51David WynnHOST
You'll want to stick around.
J
4:23Jesse Jiryu DavisGUEST
We were one of the first, actually.
J
4:27Jesse Jiryu DavisGUEST
And we actually built that system before Raft had been published in 2014.
J
4:34Jesse Jiryu DavisGUEST
And it was a home-built system that worked pretty well, but It wasn't designed with any sort of formal logic like TLA+, and it had some bugs like
C
4:48Carl SverreHOST
it showed.
S7E1 - JUXTCast S7E1: What Coding Agents Still Need From Us, with Chris Ford
C
25:07Chris FordGUEST
You might also have a finite state machine that reflects where you want the payment processing logic.
C
25:14Chris FordGUEST
even have something as fancy as TLA+, which is this modeling tool that's kind of popular in kind of hipster nerd circles, you might have various different representations of the truth.
C
25:26Chris FordGUEST
And it's my belief that the Swiss cheese model of quality applies here.
C
25:31Chris FordGUEST
So the Swiss cheese model of quality is, I've gotten and use it in aviation a lot.
9 MINS LATER
H
34:07Henry GarnerHOST
I can't say I've seen anybody actually fully committing to that way of working
C
34:12Chris FordGUEST
yet.
H
34:14Henry GarnerHOST
But presumably, well, and it's linked to the point you made about, I think it was nerds and TLA plus.
C
34:19Chris FordGUEST
Yeah.