Joe retweeted
Deciding how to help non-expert humans engage with formal verification specifications seems like one of the more important non-verifiable UX problems facing the rise of formal verification. 🧵
20
25
264
31,023
Joe retweeted
me: kinda fucked up how many people are offloading their whole lives to muse or instinct or grokbot or whatever
claude: yeah
claude: yeah
112
462
15,830
371,604
Joe retweeted
So the log_softmax bug arxiv.org/abs/2610.31642 that's been making the rounds. TorchLean also had the same problem (fixed in github.com/lean-dojo/Torc…). How come? This example is pretty illustrative of the proper mindset you need when working on formal verification. 🧵
13
35
340
37,337
The MTA has incredible time travel powers. 15 minutes ago, the next R train came in 9 minutes. Now it comes in 5 minutes!
2
0
146
4,064
Joe retweeted
At the PyTorch conference today and tomorrow, come and say hi!
We have lots of swag!
We have lots of swag!
11
10
412
16,979
Joe retweeted
After 5 months, 385 citations, and 50 pages I’m excited to finally share the preprint of the survey paper my collaborators and I have been working on:
sainati.pl/gpu-survey-pre…
We wanted to put it on arXiv but tragically they don't take survey papers anymore.
sainati.pl/gpu-survey-pre…
We wanted to put it on arXiv but tragically they don't take survey papers anymore.
2
7
32
2,255
Just landed at SFO. Will be at SPLASH this week! I’m here in my dual capacities as (a) co-author of “Fail Faster: Staging and Fast Randomness for High Performance PBT”, and (b) employee of the tee shirt company!
3
0
51
1,222
Home from SPLASH. Sorry for all the folks I wanted to catch up with but didn’t! It was a 3 day trip and I was sort of sick for the first 1.5 days :(
But saw a lot of fun stuff in the final 1.5 days.
But saw a lot of fun stuff in the final 1.5 days.
1
0
0
43
Joe retweeted
> i ask Claude if the program is verified
> he says yes
> i ask if he verified the program or a smaller, better-behaved program he imagined
> he laughs and says “it’s verified sir”
> i use the program
> it has a bug
> he says yes
> i ask if he verified the program or a smaller, better-behaved program he imagined
> he laughs and says “it’s verified sir”
> i use the program
> it has a bug
I used Opus 5.5 to formally verify the Claude Agent SDK using Lean. A couple short prompts = 16 PRs fixing various bugs and race conditions. Video attached.
TLA+ also works well. I sometimes combine Lean and TLA+ to look for issues around data flow, concurrency, and state mgmt.
nguage well, but Claude is excellent at both. This approach is super useful for formally modeling your code and finding bugs that a human probably wouldn't have spotted.
Is formal verification the future of coding (or at least, bug finding)?
16
24
586
26,299
Joe retweeted
100 problems is nothing, in programming languages we routinely solve exactly the next 700 problems
1
4
56
2,487
Unbelievably cool job
We're hiring developers to work on Hegel (property-based testing for ~every language) in London with me, and are struggling a bit to source candidates, so I wrote a less professional post about why this is the coolest job ever and you should apply.
drmaciver.com/2026/09/come-w…
1
0
11
875