Kobi Gurkan
@kobi.bsky.social
applied crypto, security, experimental things.
head of research @baincapitalcrypto.com
head of research @baincapitalcrypto.com
Want to understand ZODA? Tried to implement it and found yourself stumped?? This FAQ is just for you!
Wrote some answers for questions I’ve seen around me and those I had myself, navigating details and tradeoffs
If you have any more - let me know :)
Wrote some answers for questions I’ve seen around me and those I had myself, navigating details and tradeoffs
If you have any more - let me know :)
October 3, 2025 at 4:08 PM
Want to understand ZODA? Tried to implement it and found yourself stumped?? This FAQ is just for you!
Wrote some answers for questions I’ve seen around me and those I had myself, navigating details and tradeoffs
If you have any more - let me know :)
Wrote some answers for questions I’ve seen around me and those I had myself, navigating details and tradeoffs
If you have any more - let me know :)
While it’s not a huge increase over the baseline of no reasoning (72.8%), it still appears to work, which they see as early indication of developing steganographic abilities
See the paper here: arxiv.org/pdf/2507.02737
4/4
See the paper here: arxiv.org/pdf/2507.02737
4/4
July 6, 2025 at 8:55 PM
While it’s not a huge increase over the baseline of no reasoning (72.8%), it still appears to work, which they see as early indication of developing steganographic abilities
See the paper here: arxiv.org/pdf/2507.02737
4/4
See the paper here: arxiv.org/pdf/2507.02737
4/4
encoding it directly such that an external observer can’t understand it (“scratchpad”)
One interesting image is the one below, where with GPT 4.5, a “first-letter” encoding where first letters of sentences encoding a simple state-tracking reasoning task
3/4
One interesting image is the one below, where with GPT 4.5, a “first-letter” encoding where first letters of sentences encoding a simple state-tracking reasoning task
3/4
July 6, 2025 at 8:24 PM
encoding it directly such that an external observer can’t understand it (“scratchpad”)
One interesting image is the one below, where with GPT 4.5, a “first-letter” encoding where first letters of sentences encoding a simple state-tracking reasoning task
3/4
One interesting image is the one below, where with GPT 4.5, a “first-letter” encoding where first letters of sentences encoding a simple state-tracking reasoning task
3/4
Example result from the knowledge graph looks like this
It’s nowhere near deep enough, but I found the convenience of this process pretty amazing
One thing I want to try is doing it multiplayer, maybe with amp code or even just sharing a terminal running Claude Code
4/4
It’s nowhere near deep enough, but I found the convenience of this process pretty amazing
One thing I want to try is doing it multiplayer, maybe with amp code or even just sharing a terminal running Claude Code
4/4
June 18, 2025 at 11:11 AM
Example result from the knowledge graph looks like this
It’s nowhere near deep enough, but I found the convenience of this process pretty amazing
One thing I want to try is doing it multiplayer, maybe with amp code or even just sharing a terminal running Claude Code
4/4
It’s nowhere near deep enough, but I found the convenience of this process pretty amazing
One thing I want to try is doing it multiplayer, maybe with amp code or even just sharing a terminal running Claude Code
4/4
You get great results from Claude Code by guiding it to generate tests for itself to verify its output and supporting it by having live data and services running locally, which it utilizes since it runs in your environment
And being able to do it on your phone is the best
1/4
And being able to do it on your phone is the best
1/4
June 18, 2025 at 11:10 AM
You get great results from Claude Code by guiding it to generate tests for itself to verify its output and supporting it by having live data and services running locally, which it utilizes since it runs in your environment
And being able to do it on your phone is the best
1/4
And being able to do it on your phone is the best
1/4
Video of my talk in zksummit about Ligerito has been published!
It’s about the work by Andrija and @lmao.bsky.social introducing a small and concretely fast polynomial commitment scheme
Since then, a fun thing has happened —
1/3
It’s about the work by Andrija and @lmao.bsky.social introducing a small and concretely fast polynomial commitment scheme
Since then, a fun thing has happened —
1/3
May 28, 2025 at 8:34 PM
Video of my talk in zksummit about Ligerito has been published!
It’s about the work by Andrija and @lmao.bsky.social introducing a small and concretely fast polynomial commitment scheme
Since then, a fun thing has happened —
1/3
It’s about the work by Andrija and @lmao.bsky.social introducing a small and concretely fast polynomial commitment scheme
Since then, a fun thing has happened —
1/3
Some thoughts about how verifications in Bluesky can be extended to ZK-based methods, to achieve Verifiable Verifications
This builds on ideas from the recent verification protocol, and explores both direct integrations and lightweight ones, with different points in the tradeoff space of trust
1/2
This builds on ideas from the recent verification protocol, and explores both direct integrations and lightweight ones, with different points in the tradeoff space of trust
1/2
May 19, 2025 at 4:00 PM
Some thoughts about how verifications in Bluesky can be extended to ZK-based methods, to achieve Verifiable Verifications
This builds on ideas from the recent verification protocol, and explores both direct integrations and lightweight ones, with different points in the tradeoff space of trust
1/2
This builds on ideas from the recent verification protocol, and explores both direct integrations and lightweight ones, with different points in the tradeoff space of trust
1/2
And you can even see how it looks like in LLVM IR
3/4
3/4
April 16, 2025 at 11:36 PM
And you can even see how it looks like in LLVM IR
3/4
3/4
Native code can get as low-level as you can, with reasonable effort
2/4
2/4
April 16, 2025 at 11:36 PM
Native code can get as low-level as you can, with reasonable effort
2/4
2/4
Native code can get as low-level as you can, with reasonable effort
2/4
2/4
April 16, 2025 at 11:32 PM
Native code can get as low-level as you can, with reasonable effort
2/4
2/4
Used Gemini 2.5 to generate a game based on the “On Proving Pairings” paper by Andrija and Liam
Zap the non-residues 😂
kobigurk.github.io/zap-non-resi...
H/t to @emollick.bsky.social for the prompt
Zap the non-residues 😂
kobigurk.github.io/zap-non-resi...
H/t to @emollick.bsky.social for the prompt
April 11, 2025 at 11:53 PM
Used Gemini 2.5 to generate a game based on the “On Proving Pairings” paper by Andrija and Liam
Zap the non-residues 😂
kobigurk.github.io/zap-non-resi...
H/t to @emollick.bsky.social for the prompt
Zap the non-residues 😂
kobigurk.github.io/zap-non-resi...
H/t to @emollick.bsky.social for the prompt
April 17, 2023 at 6:03 PM