Jeremy Kun

Author of Math ∩ Programming, pimbook.org, pmfpbook.org, currently doing homomorphic encryption @ Google. Based in Portland, OR.

Public Key

npub1pe67uceztjyefem3xmphw05e65200df0hrrv60464v6w8xjtgvgqp4h36q

NIP-05 Address

[email protected]

Profile Code

nprofile1qqsqua0wvv39ezv5uacndsmh86va298hk5hm33kd86a2kd8rnf95xyqpz4mhxue69uhhyetvv9ujuerfw36x7tnsw43q83vktu

Publishing to

relay.ditto.pub

Author Public Key

npub1pe67uceztjyefem3xmphw05e65200df0hrrv60464v6w8xjtgvgqp4h36q

Show more details

Last Notes

2026-04-02 13:19:14 GMT

I think the real question about GH availability metrics is: does having zero nines actually impact their business? Or are they slacking on reliability because they can and their users won't leave.

2026-03-03 23:00:15 GMT

The ratio of CRDT tutorials online to CRDT uses in production systems is baffling.

2026-02-10 17:31:31 GMT

- reply

Also A000004 is the zero sequence.
HUGE missed opportunity. This should absolutely be A000000

2026-02-10 17:30:34 GMT

I want to be the kind of person who cites OEIS entries by id like some people cite XKCDs.
I guess that requires memorizing a lot of cool OEIS entries. Starting with A014233, the "make Miller-Rabin deterministic" sequence.
But also, A000001 is number of groups or order n. How cool is that.

2026-02-08 03:32:21 GMT

- reply

libc's math functions seems like a good source of truth; most implementations use polynomial/rational approximations for sine, cosine, erf, and friends. Some cite the Remez method for the coefficients.

2026-02-08 03:24:49 GMT

I'm doing some surveying of applications of polynomial and rational polynomial approximations for my book. Currently going through different polynomial approximation solvers looking for users that seem interesting.

2026-01-27 20:20:51 GMT

TIL if you do Miller-Rabin with the first 12 prime bases, you have a fast, exact primarily test for all 64-bit integers.
Cf. https://oeis.org/A014233 which gives the upper bound on the two-sided correctness of Miller-Rabin given that you tested on the first k bases.

2026-01-22 03:10:52 GMT

Would people find my idle musings about challenges in our group's homomorphic encryption compiler interesting? As in thoughts about features I'm actively working on

2025-12-28 15:44:56 GMT

Is there demand for resources for parents to learn how to engage with their elementary-aged children about math in a healthy way? Either helping them with schoolwork, or pre-K/Kindergarten level number sense building.
I see lots of resources on the web for teachers, curriculum/lesson plans and activity sheets, etc., but not a ton aimed at parents.

2025-12-27 02:13:18 GMT

Christmas haul was surprisingly crypto. I got a replica enigma machine (digital replica, not mechanical), and my 5yo got some invisible ink pens. Combining the two technologies may indeed turn the tide of the war.

2025-12-26 23:04:47 GMT

- reply

@nprofile…f2wy the evidence of "other equally black box tools failing to solve the same problems" is not exactly strong evidence, though we do have additional examples (beyond the recent Erdos example) of LLMs "laundering" existing literature.

2025-12-23 14:46:03 GMT

- reply

@nprofile…l4xy that's a cute one though

2025-12-18 18:05:04 GMT

Wow, TIL there was a paper from this year that achieved 250ms bootstrapping for CKKS: https://eprint.iacr.org/2025/784

2025-12-09 21:18:57 GMT

- reply

@nprofile…stct you might enjoy reading Ben Recht's newsletter, which covers these topics in detail.

2025-11-24 18:15:36 GMT

I have to admit, I'm long overdue for writing some blog posts with actual implementations of the top FHE schemes.

2025-11-18 14:50:51 GMT

- reply

@nprofile…ztfm @nprofile…63jn @nprofile…6fuz that is all well and good, though it would be a shame if "leading figures" get an exception to a policy solely on account of their stature.
I was mostly curious about the "letters to the editor" part. That is even less like "research" than a survey paper, even though I'd agree on the value of having that in a discoverable place.

2025-11-18 03:40:07 GMT

- reply

@nprofile…ztfm would these two articles no longer meet the new standards of publication on the arxiv?

2025-11-18 00:18:53 GMT

Bicyclic Matrix-Matrix Multiplication in Fully Homomorphic Encryption
https://www.jeremykun.com/2025/11/17/bicyclic-matrix-matrix-multiplication-in-fully-homomorphic-encryption/

2025-11-17 17:05:22 GMT

Another followup to the big "AlphaEvolve improves math conjectures" paper: https://arxiv.org/abs/2506.16750
100x improvement on one of the conjectures. "Our function was found using simulated annealing and gradient based methods rather than using large language models."
IIUC, the AlphaEvolve examples ran for _weeks_ to find a new solution.
Apparently it seems most of these conjectures had not been seriously attacked by computational methods before the AlphaEvolve paper.

2025-11-15 17:14:53 GMT

Noticing LLM spam on GitHub repos: any time I open an issue on my cryptography compiler project, some rando whose profile lists no relation to crypto or compilers files a PR with an LLM-generated implementation.
What to do? My security brain feels like these are scammers trying to build up legit-looking profiles for the purpose of eventually doing supply chain attacks.

2025-11-08 01:40:25 GMT

- reply

@nprofile…zlur it's your golden opportunity to give it a fun name!!

2025-10-20 15:49:00 GMT

Today I heard sort of offhand that one way people can detect AI music is to look for compression artifacts in the tracks. At first I thought, well how would you know the difference between compression artifacts that are supposed to be there versus "bad" compression artifacts introduced by an AI trained on compressed media?
Apparently the answer is that you can distinguish compression artifacts that show up in a 128kbps mp3, and if those show up in an mp3 labeled as 256kbps, then it's a signal that an AI generated it.
Does anyone have any more detailed info on this, or if there's more to these methods? Sadly it does seem like the kind of detection method that could be bypassed: just have a separate 256kbps model only trained on 256kbps tracks.
It also makes me wonder about text... what has been tried and known not to work? I suspect nothing has been found to work for real.

2025-08-22 13:03:06 GMT

- reply

@nprofile…hr6z @nprofile…dngu if you want the vicarious experience check out https://www.lostartoflogarithms.com/chapter03/

2025-08-22 00:11:20 GMT

- reply

@nprofile…dngu I get your point, but old books of logarithm tables had plenty of errors.... Error tolerance is a matter of degree, not kind

2025-07-28 15:31:59 GMT

- reply

@nprofile…qmju As many have said before me, an LLM is often like talking to yourself in a mirror.

2025-07-27 01:57:51 GMT

Do any solo devs do shareware or similar for things like developer tools?

2025-07-18 21:16:09 GMT

Frequently asked questions about FHE
https://www.jeremykun.com/frequently-asked-questions-about-fhe/

2025-07-09 20:00:20 GMT

I found one of those 4-digit number clickers and showed it to my toddler. He has been playing with it and using it to count people who walk past our house.
We use it to talk about place value, the idea of offloading our memory to a physical device (I keep pretending I forget the current number and he is delighted to remind me by showing me the clicker), and he learned about "overflow", which he describes as "when the number gets so big that the computer gets tired." I told him that it's like if you filled up a cup too much and it overflowed, but then instead of staying full the cup suddenly became empty.
My son also asked: how does this computer work without any energy? (electricity) And I realized that yes, this is a computer.
https://media.mathstodon.xyz/media_attachments/files/114/825/036/893/578/126/original/b75c481a6099e75c.png

2025-03-17 22:39:02 GMT

- reply

@nprofile…59k6 for prototyping shell pipelines, I often use https://github.com/akavel/up there were many cases I had where the commands I wanted to run were not easy to do in vim, e.g. a jq query or ask script. This let me easily prototype those live with the data

2025-03-12 23:04:48 GMT

Seems there are lots of message requests on Twitter that I just never see. Sucks I guess.

2025-02-26 06:39:40 GMT

- reply

@nprofile…59k6 what was the impact of that "noted failure" for you personally? Did you have to pay a fine or do extra work or something?

2025-02-03 23:13:57 GMT

TIL this git command searches the history of a repo for a string (including the first commit that introduced the string)
git log -S <search string> --source --all
Useful for finding the origin of an error message that has moved around the codebase. h/t https://stackoverflow.com/a/5816177/438830

2024-12-16 17:33:51 GMT

Ada Lovelace's famous "note G" algorithm to compute Bernoulli numbers apparently contained a bug (minor, typographical in nature).
ONE OF US. ONE OF US. ONE OF US.

2024-11-30 05:28:28 GMT

- reply

@nprofile…gyrf I think you'd see it within the first ten minutes of https://www.youtube.com/watch?v=7ofnlCFzlqg
It's this llvm::TrailingObjects thing, though I don't understand it well, and on a quick skim I don't see them applying this to an entire basic block. It does seem like the goal here is to avoid pointer chasing either way.

2024-11-30 05:23:24 GMT

- reply

@nprofile…gyrf I recall a talk from the MLIR folks about their internal memory layouts that seemed something like this, though I don't remember all the details and I don't think they tried compiling with GPU acceleration.

2024-11-30 05:12:57 GMT

- reply

@nprofile…gyrf Just curious, what is the core idea as to how to avoid pointer chasing?

2024-11-25 03:46:49 GMT

@nprofile…749s did you ever settle on a shapeways replacement? People are asking me what they should get me for Christmas and I always put some of your stuff on the list (the ones I haven't already got)

2024-11-22 16:19:57 GMT

There's a bulletin board at my local bagel shop, on it is an advertisement for ABACUS LESSONS and every single phone number slip has been taken.

2024-09-27 02:50:40 GMT

Just a few finishing touches left, but I'm pretty happy with it.
https://media.mathstodon.xyz/media_attachments/files/113/207/239/129/813/357/original/fefe0e1824734fdc.jpg

2024-09-02 21:27:01 GMT

I wrote up a little article about a problem that shows up in homomorphic encryption called "cheapest shift network" https://www.jeremykun.com/2024/09/02/shift-networks/

2024-08-10 23:00:47 GMT

File Over App
https://stephango.com/file-over-app

2024-08-02 22:50:14 GMT

Dad achievement unlocked: broke a drill bit.

2024-07-25 17:18:26 GMT

- reply

@npub14fs…qegk no sir, they're physical games 😭😭😭😭😭😭😭😭

2024-07-25 17:03:14 GMT

I have gotten into a habit of making my own games (either copying someone else's game or trying to make a new game) but I have nobody to play/test with 😭

2024-06-21 19:33:42 GMT

Always inject whimsy where it isn't expected

2024-05-17 21:06:58 GMT

Rounding is a bitch
>>> np.sin(-1.57079633)
-1.0
>>> np.sin(-1.57079632)
-1.0

2024-03-29 13:25:41 GMT

The downside of having a stimulating and exciting conference week is that my book writing schedule is molding over in the corner.

2024-03-19 17:51:02 GMT

Heading to Toronto and I'll have a day to explore before conferences ramp up. Looking for recommendations, particularly stuff that's not typical tourist stuff. Quirky/unique shops, local distilleries, local artists, I like that kind of stuff.

2024-03-14 16:50:07 GMT

"The insatiable hunger of OpenAI"
https://wimvanderbauwhede.codeberg.page/articles/the-insatiable-hunger-of-openai/
It's ironic how all the existential risk people warn us with a parable of a "paperclip optimizer" and then turn around and claim we should funnel society's resources into the current wave of AI.
And what's it all for? To teach AIs to farm for resources in video games. To enable disinformation at scale. To have robots write and read our pointless emails. To take the humans out of art. To make the extractive machinery of our society more extractive.
The author ends with "the world can't afford this growth in AI." Indeed, and we're not getting much for what we're paying.

2024-03-14 16:31:58 GMT

Paperclip optimizer? How about running the climate into the ground so we can teach AIs to farm resources in online video games.

2024-02-03 15:22:59 GMT

The "sweet spot" chapter outline I've settled into while writing my book is:
- 2 pages of social/problem context
- 2 pages of math
- 1 page relating the math to the problem context
- 1 page of demo code

2023-12-15 14:36:44 GMT

- reply

@npub16uz…xx07 oh god now I can't close your website!!

2023-09-03 05:42:41 GMT

How have I never heard of the group-ring of a group until now??? https://en.m.wikipedia.org/wiki/Group_ring

2023-08-19 19:57:12 GMT

It's Saturday Morning Math Writer's Jam!
This week I'm looking at applications of math to wildlife conservation. I had an excellent interview on Friday with Lily Xu, a PhD student at Harvard working on the front lines in places like Uganda, Cambodia, and Belize.

2023-07-09 20:50:34 GMT

- reply

@npub1dnn…yyg8 @npub1hyk…3p33 there's got to be a T-shirt that says seize the means of computation