Vikram Seraph 🪽

Vikram Seraph 🪽

@vikramsaraph.com

Software engineer, AI/ML researcher, and mathematician at Johns Hopkins APL. Former New Englander, current Marylander. Brown CS PhD and Notre Dame math alum. Nerd of sorts (computers, math, language, puzzles, games, books, music). Opinions are my own.

1.8K followers741 following2.3K posts

Vikram Seraph 🪽 reposted

Sam Wang
@samwang.bsky.social

House and Senate trends, moving together. The Princeton Election Consortium is live: election.princeton.edu The Senate and House meta-leads are moving together over time, suggesting that a tide of Democratic opinion driving parallel changes in both datasets.

Vikram Seraph 🪽
@vikramsaraph.com

I tried this again 6 months later. Claude Code got *way* farther this time (though I did use Opus 5.x): github.com/vhxs/simplic... I have the simplicial approximation theorem. The definitions are compatible with already exists in Mathlib. It took shockingly little input from me...

github.comGitHub - vhxs/simplicial_complexesContribute to vhxs/simplicial_complexes development by creating an account on GitHub.
Vikram Seraph 🪽@vikramsaraph.com

Sonnet 4.5 got surprisingly far with proving the simplicial approximation theorem in Lean. I was trying to define combinatorial simplicial complexes in such a way that the geometric realization functor uses the existing definition of geometric complexes already in Lean’s mathlib.

View full thread
Vikram Seraph 🪽
@vikramsaraph.com

I tried this again 6 months later. Claude Code got *way* farther this time (though I did use Opus 5.x): github.com/vhxs/simplic... I have the simplicial approximation theorem. The definitions are compatible with already exists in Mathlib. It took shockingly little input from me...

github.comGitHub - vhxs/simplicial_complexesContribute to vhxs/simplicial_complexes development by creating an account on GitHub.
Vikram Seraph 🪽@vikramsaraph.com

Sonnet 4.5 got surprisingly far with proving the simplicial approximation theorem in Lean. I was trying to define combinatorial simplicial complexes in such a way that the geometric realization functor uses the existing definition of geometric complexes already in Lean’s mathlib.

View full thread
Vikram Seraph 🪽
@vikramsaraph.com

I tried this again 6 months later. Claude Code got *way* farther this time (though I did use Opus 5.x): github.com/vhxs/simplic... I have the simplicial approximation theorem. The definitions are compatible with already exists in Mathlib. It took shockingly little input from me...

github.comGitHub - vhxs/simplicial_complexesContribute to vhxs/simplicial_complexes development by creating an account on GitHub.
Vikram Seraph 🪽@vikramsaraph.com

Sonnet 4.5 got surprisingly far with proving the simplicial approximation theorem in Lean. I was trying to define combinatorial simplicial complexes in such a way that the geometric realization functor uses the existing definition of geometric complexes already in Lean’s mathlib.

Vikram Seraph 🪽
@vikramsaraph.com

... though there were some important decision points where I had to provide input. I didn't have to fight CC to break down proofs into manageable lemmas, it did that all on its own (last time it would blow through my token budget if I wasn't careful)

Vikram Seraph 🪽 reposted

Clément Canonne
@ccanonne.github.io

OpenAI has claimed a refutation of the Exponential Time Hypothetis (ETH). Russell Impagliazzo, Mohan Paturi and Ryan Williams (@rrwilliams.bsky.social) have been asked to organize a community-led effort to evaluate this claim. More here: docs.google.com/document/d/1... h/t @gautamkamath.com

docs.google.comMessage to the TCS Community on ETHDear colleagues, OpenAI has informed us that it has obtained a claimed refutation of the Exponential Time Hypothesis and plans to release the supporting documents to the community. Our names may appea...
Vikram Seraph 🪽
@vikramsaraph.com

This happens tomorrow: snfagora.jhu.edu/event/the-ho... > The debate will be recorded live and shared with global audiences through Open to Debate’s YouTube channel, podcast feeds, Substack, and more than 250 NPR stations.

snfagora.jhu.eduThe Hopkins Forum: Will Generative AI Enhance or Stunt College Learning?ABOUT THE DEBATE: Like it or not, AI has entered the chat and nowhere is it eliciting more decidedly mixed feelings than on college campuses. Research shows that the majority […]
Vikram Seraph 🪽
@vikramsaraph.com

I've had GLM 5.3 write CUDA kernels via OpenCode for a thing I'm going to intentionally be vague about and... wow am I impressed. Not that I don't want to understand for myself the code it's writing, but it's correct *and* performant.

Vikram Seraph 🪽 reposted

rain 🌦️
@sunshowers.io

Update: Nick G. from Anthropic has responded on HN, and I believe that the attribution is now sufficient. I stand by my posts as of the time I wrote them.

news.ycombinator.comHi! I am the original author of the blog post and wanted to clear the air a bit ... | Hacker News
rain 🌦️@sunshowers.io

Update on yesterday's post about Anthropic ripping off Mononoke:

sunshowers.ioMore on Anthropic's Scion ripping off MononokeAnswering questions and concerns.

Vikram Seraph 🪽 reposted

Ted Underwood
@tedunderwood.com

If you're wondering about the genes that went into creating "Close All Your Windows," a bunch of them (esp manic pixie Windows XP) have to be from this:

This post is unavailable.

Vikram Seraph 🪽 reposted

Sung Kim
@sungkim.bsky.social

Can you develop a working application with an AI coding agent in a matter of hours? Yes. Can you get it production ready in a matter of hours? Usually not.

Vikram Seraph 🪽 reposted

Matt Kane
@mk.gg

Cloudflare is becoming a free certificate authority, and will be the first to issue post-quantum Merkle Tree Certificates

blog.cloudflare.comBuilding a certificate authority for the whole InternetTwelve years after launching Universal SSL, Cloudflare is applying to become a certificate authority.
Matt Kane@mk.gg

Whoa I just had a peek at the announcements that we're putting out today, and one in particular is MASSIVE. Watch this space in 20 minutes. ⚛️

Vikram Seraph 🪽
@vikramsaraph.com

Will post a link to the proceedings on IEEE Xplore once it’s up. This work was done Q1-Q2 ‘24 before standardization of terminology and protocols for agentic AI really took off. I’ll say more when I have a paper online to point at

Vikram Seraph 🪽 reposted

Andrew Feeney
@andrewfeeney.au

Someone yelled across the gym this morning asking if I could spot them. I said “That’s a load bearing observation and it’s worth calling out.”

Vikram Seraph 🪽
@vikramsaraph.com

Last travel related thing. I have a friend who is about to get their fourth citizenship (among them US, all of them there are legit reasons for having gotten). Made me realize that the set of sets of compatible citizenships form a simplicial complex (some are not compatible with others).

View full thread
Vikram Seraph 🪽
@vikramsaraph.com

Here’s the log: - Walk from Airbnb in Girona to train station - Take train to Barcelona train station - Take taxi to BCN - Fly to BOS - Drive to extended family’s house in the area because whoops flight cancelled. Spend night there, flight rebooked for next day

Vikram Seraph 🪽
@vikramsaraph.com

- Wake up next day, whoops flight cancelled again, better just take the train, book train ticket - Drive to South Station - Take train all the way to Washington Union Station - Take metro to Arlington - Walk to parking garage - Drive home Did I mention we have a toddler