Yaron (Ron) Minsky

Occasional OCaml programmer. Host of @signalsthreads.

@[email protected]
@yminsky.bsky.social

[Yaron (Ron) Minsky @yminsky]

Totally agreed. One direction I'm excited to try is to see what happens as we make lightweight specifications a thing that you can write down in the signature of an individual module. The direction there is rather than trying to specify a system from the top down, you'd slowly build in more specifications and properties from the bottom up. I'm very curious to see what people do with that kind of system.

[Yaron (Ron) Minsky @yminsky]

@charliermarsh Deep cut

[John Crepezzi @seejohnrun]

As a long time Signals and Threads listener, I was really excited to have this opportunity to talk about the work our team has done over the past ~4 years. Hope you enjoy 😊

[Yaron (Ron) Minsky @yminsky]

@difficultyang Great use for formal proofs?

[Yaron (Ron) Minsky @yminsky]

Come for the agentic coding, stay for the discussion of QBasic, Ruby on Rails, and maintainability of large dynamically typed codebases.

[Yaron (Ron) Minsky @yminsky]

A new Signals and Threads, with the great @seejohnrun, talking about AI and software engineering, and how the world is changing under our feet...

[Yaron (Ron) Minsky @yminsky]

The actual post is here.

https://t.co/AGV3hUHZTF

[Sam Tobin-Hochstadt @samth]

The responses to this are so wholesome it's made my whole day.

[Yaron (Ron) Minsky @yminsky]

@mbacarella That's a little disappointing. Most of these sounds kinda dumb.

[Yaron (Ron) Minsky @yminsky]

I'm desperately curious to know what the 20% of me "doing my thing" consists of....

[Shriram Krishnamurthi (primary: Bluesky) @ShriramKMurthi]

Signals and Threads episodes are bangers, but this may be the first to become part of my PL course's "text". About 50% is a compressed version of my GC lectures, about 30% is modern updates, and 20% is @yminsky doing his thing w/ Stephen Dolan. (-:
https://t.co/iaqIPsWEiR

[Yaron (Ron) Minsky @yminsky]

@ShriramKMurthi That was a fun bit of the interview. I really learned a lot on that one! That's also really the first time I understood why stack allocation is so much more efficient than allocating on the minor heap.

[Shriram Krishnamurthi (primary: Bluesky) @ShriramKMurthi]

Signals and Threads episodes are bangers, but this may be the first to become part of my PL course's "text". About 50% is a compressed version of my GC lectures, about 30% is modern updates, and 20% is @yminsky doing his thing w/ Stephen Dolan. (-:
https://t.co/iaqIPsWEiR

[Yaron (Ron) Minsky @yminsky]

@JulesJacobs5 @JAldrichPL It's a weirdly high-stakes bit of philsophy.

[Yaron (Ron) Minsky @yminsky]

@JulesJacobs5 @JAldrichPL I worry that we're purposely teaching LLMs to act as if they might be conscious, in a way that will confuse us into concluding they are. And that confusion in its full form would put a heavy (and arguably wrong!) thumb on our moral calculus.

[Yaron (Ron) Minsky @yminsky]

@JulesJacobs5 @HerrDreyer @ArjunGuha @krismicinski Which raises the interesting question of whether models are presently good at checking proofs. I wonder if there are any benchmarks of this kind out there.

[Yaron (Ron) Minsky @yminsky]

@JulesJacobs5 @HerrDreyer @ArjunGuha @krismicinski My guess would be that RL with lean as the validator is a big part of how the models get good at math, whether or not they use lean for a given proof.

[Yaron (Ron) Minsky @yminsky]

I'm so here for this.

[Jules Jacobs @JulesJacobs5]

The OCaml module system is perfect for formal specifications of a module's implementation.