Sign in

sasha

@sasha.place
159 followers 196 following 174 posts

PhD Student at UMD studying Cryptography website: sasha.place Formerly at Meta/Cornell

PostsRepliesMedia
sasha @sasha.place · 17/09/2026
eprint.iacr.org/2026/2053 My first research paper from my internship at a16z is up online! We made encryption schemes for encrypted mempools better!
eprint.iacr.org
Weighted Batch Threshold Encryption with Efficient DKG
Batch threshold encryption allows a committee to decrypt a selected batch of ciphertexts using one short pre-decryption key per party, while ciphertexts outside the batch remain private. This makes ba...
210
sasha @sasha.place · 13/09/2026
000
sasha @sasha.place · 11/09/2026
going up for tenure in the UK they ask you what your haitch-index is
000
sasha @sasha.place · 08/09/2026
another major AI tell I have noticed recently: AI loves writing sentences with semicolons in proofs. I almost never see these in human written papers
000
Reposted by sasha
ePrint Updates @eprint.ing.bot · 03/09/2026
Private and Verifiable Outsourcing of Open-Weight LLM Inference (Kanav Gupta, Jonathan Katz, Ian Miers) ia.cr/2026/1849
Abstract. Open-weight models allow clients to run LLMs locally, thus keeping their data private from untrusted providers. However, running large models requires massive hardware and storage resources (especially challenging on resource-constrained devices like smartphones), limiting local execution to smaller models. This leaves clients with a frustrating compromise: settle for a less-capable model that can be run locally, or sacrifice privacy by sending queries to an external server.

We present an efficient protocol that allows a client to privately and verifiably outsource LLM inference of an open-weight model to a pair of malicious (but non-colluding) servers. Privacy implies that neither server learns anything about the client’s queries. At the same time, the client can verify the claimed result using information posted by the model owner along with the model weights.

Compared to prior state-of-the-art for private LLM inference (SIGMA, PETS’ 24) – which does not provide verifiability – our protocol is $11 − −14$ faster while imposing no overhead at the servers (beyond the cost of inference in the original model). Our protocol also scales to larger models not supported by prior work: for example, with our protocol a client can run the Llama 2-70B model using just 179~MB of local storage (instead of the 140~GB required to run the model locally).
Image showing part 2 of abstract.
032
sasha @sasha.place · 03/09/2026
love the sentence in the Eurocrypt Call for Papers saying they're not gonna give out best paper awards if all the papers are bad.
010
sasha @sasha.place · 21/08/2026
I've been using AI reviews to improve my papers and it is remarkably nitpicky sometimes. AI is really good at asking for a bunch of random low level implementation considerations or a bunch of extra experiments to run.
100
sasha @sasha.place · 19/08/2026
how does IEEE have so many really bad journals/conferences? what leads to this?
000
sasha @sasha.place · 19/08/2026
000
sasha @sasha.place · 18/08/2026
Has anybody ever heard of this zk-SNARK library??? github.com/zkonduit/ezkl I am kind of surprised that it has 1.2k stars and I have never heard of it or seen it used anywhere.
github.com
GitHub - zkonduit/ezkl: ezkl is an engine for doing inference for deep learning models and other computational graphs in a zk-snark (ZKML). Use it from Python, Javascript, or the command line.
ezkl is an engine for doing inference for deep learning models and other computational graphs in a zk-snark (ZKML). Use it from Python, Javascript, or the command line. - zkonduit/ezkl
120
sasha @sasha.place · 17/08/2026
i met nicholas carlini and saw adam sandler in the last 3 days :D
000
sasha @sasha.place · 14/08/2026
today's feeling
010
sasha @sasha.place · 14/08/2026
I'm at USENIX Security this week. A pretty large number of authors are having other people give their presentations because of visa issues. This morning I watched one of these substitute presenters give a presentation in the third person ("they did x", "they found this result")
000
Reposted by sasha
ePrint Updates @eprint.ing.bot · 06/08/2026
zk-Cinema: Proving Video Provenance in Zero Knowledge (Alexander Frolov, Jianfeng Guo, Xinyi Zhao, Trisha Datta, Dan Boneh, Ian Miers) ia.cr/2026/1598
Abstract. Video provenance is an important problem on the modern internet. In response, the Coalition for Content Provenance and Authenticity (C2PA) has developed a standard for verifying video and image provenance where cameras sign captured videos with an on-device secret key. Since videos are generally edited and resized before be- ing posted, the C2PA signature from a camera cannot be used as is to verify provenance of published videos. Prior work has developed zero-knowledge techniques for verifying provenance of edited im- ages and videos. In this work, we develop new efficient techniques for producing such zero-knowledge proofs. First, we show how to represent common video edits as matrix multiplications in a form that is particularly friendly for zero-knowledge provers and enables a number of optimizations. Second, we develop a SNARK-friendly video representation, which we call sfvr, that reduces prover work for video editing. Third, we design new efficient methods for incor- porating signed data into a SNARK proof. To evaluate our designs, we built an end-to-end system for proving edits to a signed video. In our end-to-end system, we optimize the NeutronNova folding scheme for high-arity folding. To scale the size of our Neutron- Nova proofs, we implement a “Read-Write Streaming” version of NeutronNova to take advantage of high-performance storage and parallel computing resources. Our system achieves competitive performance and scale relative to prior work.
Image showing part 2 of abstract.
032
sasha @sasha.place · 06/08/2026
New preprint! a bunch of improvements and cool techniques for better zero-knowledge proofs of image/video editing. Will write more about it soon. Read here: eprint.iacr.org/2026/1598
031
sasha @sasha.place · 05/08/2026
I ran a GPT 5.6 typo check on this paper (26 pages of dense 2 column text) and it took me like 90 minutes to fix all the typos/grammars/minor math things 😬 made me feel unbelievably dyslexic.
000
sasha @sasha.place · 05/08/2026
waiting for the eprint.iacr.org admins to post my submission
000
sasha @sasha.place · 30/07/2026
current academic writing problem: I've reviewers jump down my throat a few times over what I thought were minor writing mistakes. Now, any time I'm writing something which will likely get scrutinized by reviewers, I get really bad writer's block because im thinking about every way they could be mean
200
sasha @sasha.place · 29/07/2026
one of these weird academic writing things that a lot of my coauthors do is use the word "crucially" way too much when writing the introductions to papers. i do not like the way this sounds. are there people out there who see the word "crucially" in academic writing and say "oh boy I like this now"
000
sasha @sasha.place · 21/07/2026
2 more AI tells in AI generated papers I've seen a lot recently: 1) When AI's use a mathematical concept named after somebody (e.g. Shamir Secret Sharing), they start referring to it by just the last name and nothing else, which leads to a lot of weird and hard to read writing.
100
sasha @sasha.place · 15/07/2026
two AI writing tells in AI generated research papers I have noticed recently which I think aren't sufficiently discussed: 1) Unnecessarily hyphenating words a lot. 2) Using the word caveat a lot.
020
sasha @sasha.place · 09/07/2026
more crypto papers should start with sentences that read like Star Wars scrolling text
0211
sasha @sasha.place · 20/06/2026
coding theory is my passion
130
sasha @sasha.place · 27/05/2026
random cryptography pet peeve: not everything needs syntax and definitions! multiple papers in the "image/video editing with zk-SNARKs" literature make cryptographic definitions for what it means to edit a video. the proof is then basically "it reduces to the security of the underlying SNARK".
120
sasha @sasha.place · 21/05/2026
AI assistants are still a pretty recent phenomenon and it will take some time for rules around etiquette to develop. however, if i ask you a question and your response is to type it into an AI chatbot and send me the response, I think that is disrespectful.
110
sasha @sasha.place · 11/05/2026
i've been adding claude code into my workflow a lot recently, it is super interesting how much faster it is for writing simple Python code than for writing stuff kind of deep into a Rust project.
430
sasha @sasha.place · 29/04/2026
it brings me great pleasure to announce that i found a codebase that copies this typo from this paper
020
sasha @sasha.place · 24/04/2026
they call me 007: - 0 published conference papers - 0 published journal papers - 7 rejected submissions (these are my real current numbers in grad school)
110
sasha @sasha.place · 23/04/2026
fun stuff from digging around on eprint.iacr.org : Jens Groth updated a paper from 2009 this week.
110
sasha @sasha.place · 13/04/2026
I recently spent some time reading "Unlocking the lookup singularity with Lasso" (eprint.iacr.org/2023/1216) and the sequel papers. I think TaSSLE (eprint.iacr.org/2024/1075) is the best companion/exposition work for understanding Lasso of the ones out there.
eprint.iacr.org
Unlocking the lookup singularity with Lasso
This paper introduces Lasso, a new family of lookup arguments, which allow an untrusted prover to commit to a vector $a \in \mathbb{F}^m$ and prove that all entries of a reside in some predetermined t...
200
sasha @sasha.place · 09/04/2026
"keyses"??? did gollum name the variables in this paper?
000
sasha @sasha.place · 09/04/2026
new blog post! The USENIX Security '26 Call for Papers has examples of the required Ethics Appendix which they think are well written. If you read the example ethics appendix for cryptography, it is actually somewhat strange: www.sasha.place/blog/2026/04....
sasha.place
About the example ethics appendices in the USENIX Call for Papers
TL;DRThe USENIX Security ‘26 Call for Papers has some examples of papers with well-written ethics appendices, which are now required to submit to the confere...
100
sasha @sasha.place · 02/04/2026
i just got a cold email offering 70$ an hour for a "tutor" role where I would mentor a high school student on a research project to submit to high school science fairs. crazy stuff
010
sasha @sasha.place · 19/03/2026
more complaining about paper reviews: I have now twice gotten the paper review comment "can you do a tight, concrete security reduction?" for protocols using zk-SNARKs as a subcomponent. No paper in the applied SNARK literature does this! it's messy!
100
sasha @sasha.place · 17/03/2026
performative zk-SNARKs 🙄 wearing tote bags and stuff
100
sasha @sasha.place · 09/03/2026
New preprint is up! We put it up so we can submit to poster sessions, I will write more about it when it gets accepted.
010
sasha @sasha.place · 18/02/2026
credit where credit is due: this AI generated code is the only research-grade cryptography code I have ever seen that tried to properly do domain separation.
010
sasha @sasha.place · 01/02/2026
typod zero-knowledge as "aero-knowledge". we're going aero knowledge baby 😎
010
sasha @sasha.place · 30/01/2026
adventures in vibe coding cryptography: I was asked to review some AI generated cryptography code. It implemented a multiset hash function which computed the hash of a multiset by hashing each element individually with Poseidon, and then adding them together as field elements.
110
sasha @sasha.place · 27/01/2026
a unique winter sight in College Park: a common sledding location is the hill next to the IONQ office :)
000
sasha @sasha.place · 27/01/2026
I have to work on a project for grad school which uses the RISC-Zero zkVM. Reiterating a hot take I have had previously: zkVMs are not actually that developer friendly as soon as you need to try to optimize your code in any way!
sasha.place
Trade-offs and Pitfalls in zkVM design (or, some ways to make your zkVM code 3-10 times slower)
In the past few months, I have done some work writing optimized code for the Succinct SP1 zkVM. Tuning zkVM code is a remarkably fruitful area for finding wa...
100
sasha @sasha.place · 20/01/2026
There was a small proof I was stuck on, so I asked a few different AI assistants to help me prove it. They were all wrong, but ChatGPT did give a correct outline of a proof, where it was correct after working out the details.
000
sasha @sasha.place · 16/01/2026
i got rejection #4 of grad school today! I got a new insane reviewer highlight: Our paper says "this protocol can be made noninteractive via the Fiat-Shamir transform" in a few places, and the reviewer comments "if you don't pass everything into the F-S hash function, this would be insecure".
100
sasha @sasha.place · 15/01/2026
applied cryptography take: when a lot of cryptography papers need a publicly verifiable, append-only database, they almost always write "blockchain," when I think they should be writing "blockchain/transparency log". there are rarely economic incentives that would make a public blockchain make sense
110
sasha @sasha.place · 14/01/2026
i have spent a really frustrating amount of time in grad school re reading stuff I wrote and asking "how could a reviewer misread this and use this as an excuse to reject the paper without reading the rest of it"
110
sasha @sasha.place · 07/01/2026
I have used this simple technique to find issues in 3 of the 4 papers I have reviewed in grad school so far: When reading a proof/protocol, just ask yourself "is every quantity that needs to be polynomially bounded for this to work actually poly bounded?"
100
sasha @sasha.place · 24/12/2025
i finished my last class of grad school! i am done with classes for life! (hopefully)
000
Reposted by sasha
Nick Kapur @nickkapur.bsky.social · 02/11/2025
"Some random background character from Les Misérables" x.com/dailyportalz...
2785113
Reposted by sasha
Nick Kapur @nickkapur.bsky.social · 02/11/2025
Japan's "Mundane Halloween" costume contest is back! Each year website DailyPortalZ holds a contest where people dress up as something super duper ordinary. Here's a thread of some of my favorites from the 2025 contest! #MundaneHalloween
3048032809
sasha @sasha.place · 08/12/2025
our paper does client-side SNARK proving, and this AI generated response asked why we didn't use the cq lookup argument for a lookup table with 2^32 elements. cq involves storing an elliptic curve point per element of your lookup table! (2 if you count the KZG setup).
110