Founder @ ParaQuery (YC P25) | Data processing for the GPU era. https://xitter.cf/t.co/tWz7U2i0pl Dilettante of FP, PL, physics, quantum computing, and trolling AI with math.
Joined June 2011
- Tweets1K
- Following903
- Followers428
- Likes3.2K
Win Wang retweeted
The Scala Summit 2026 recordings are out! 🎥
All 11 talks from our October 19 conference in San Francisco: AI and agents, capture checking, Chisel, ZIO, GPUs for type inference, and more.
watch.softinio.com/w/p/1y91H…
#Scala #ScalaSummit #BayAreaScala #SFScala #FunctionalProgramming
cc @BayAreaScala @sfscala @Winium @higherkindedtpe @adamwarski @lukasz_bialy @RomanowskiKr @liosedhel @VirtusLab @scala_lang @tstokrk
"On the (allegedly) sequential nature of unification"...
Type inference is not sequential in practice. Measured STLC on a GPU:
Win Wang retweeted
My favorite talk from Scala Days 2026 was about baking refinement types right into the compiler.
Unlike libraries like Iron or Refined, it adds a built-in solver that verifies logic and math across multiple variables at compile time with zero runtime cost.
Moving from simple checks to full verification is how Scala's type system should evolve to constrain LLMs - AI code won't compile unless the math holds up.
Win Wang retweeted
🎤 Speaker spotlight: Win Wang @Winium , founder of ParaQuery, ex-Twitter distributed builds & parallel Scala compilation.
"GPUs for Type Inference and Typeclasses": we love strong types, we hate compile times. Can we throw a GPU at implicit resolution?
🗓️ Mon Oct 19 · SF · 6 days to go
🎟️ Get your ticket: 2026.scalasummit.com
#Scala #Scala3 #BayAreaScala #SFScala #SanFrancisco #ScalaSummit2026
ALT picture of speaker
I love it when LLMs matter-of-factly roast their own code, as if they weren't the perpetrator. Just something funny about it.
"The data shows these lists are massively oversized relative to actual usage, wasting 98%+ of their allocated capacity..."
Win Wang retweeted
While I was dreaming today, a solution to proper agents sandboxing was revealed to me.
Let’s put their sandbox inside another sandbox.
This way when the agents escape the first sandbox, they think they’re free and stop trying.
Opus 5.5 is so good at making mobile-friendly PDFs... with dark theme! Great to actually read technical concepts on the go.
Win Wang retweeted
You can have bad days.
And then you have even worse days.
And you realise bad days were good days.
There's something so interesting about having agents autonomously run performance experiments on a home server/workstation. You can hear the fans ramp up. Takes me back to the days when you could hear the HDD seeks.
Yeah, types can go faster if you throw a GPU at them.
winwang.blog/posts/dphm-1
I wanted to see (with AI) how much math would be needed to get a "cup" to tilt in a way which makes the waterline parabolic/hyperbolic instead of elliptic. AI loves to cheat on this idea (no you can't use a cone cup), but is now currently exploring transfinite abstract cups 😅
It has never been easier to write decent-performance code. Just make the intern (uh, AI) inspect your PTX/bytecode/IR and pull up relevant docs (*for compute-bound programs)
Overheard future math aphorism: "RH is false, but its practical failure rate is provably O(1/ε). Since its falsity has already impacted us once, a second independent event would be O(1/ε^2), so we may safely continue assuming RH."
Have you ever:
1. asked AI to try and solve some problem
2. have it come back with an unfinished answer
3. get annoyed because it yet-again "failed" the task
4. then remember that it was a open research problem?
Oops.
1. Discuss technical changes with AI, get to an agreement, have it implement.
2. Review code, ask it why the hell it didn't do the thing we agreed on.
3. "To conserve existing behavior" -- despite agreeing that the change has *provably* the same semantics.
... "Alignment" 🤦
Win Wang retweeted
Fun programming history fact: @akka_io_ was Lightbend was Typesafe before it dropped Scala and Play and focused on Akka. Lightbend was a name that replaced Typesafe because they thought it's a bad name for marketing. And now the hottest thing is @typesafeai. And I built TypeSec: pair.rs/typesec.