Hacker News: syumei

New comment by syumei in "Claude Code and Codex can have real-time conversation via Git"

syumei — Thu, 04 Jun 2026 18:08:58 +0000

Fair criticism. I don’t think the value is just "agents can send text to each other”; that part has many implementaton and design choices. The value of h5i I’m exploring is making the intermediate state reviewable: review requests, risks, handoffs, unresolved claims, associated prompts and AI-to-Ai conveersation, and final decisions tied to the branch/PR.

Claude Code and Codex can have real-time conversation via Git

syumei — Sun, 31 May 2026 14:12:48 +0000

Article URL: https://medium.com/@Koukyosyumei/claude-code-and-codex-can-have-real-time-conversation-via-git-f95b696c1c05

Comments URL: https://news.ycombinator.com/item?id=48345837

Points: 103

# Comments: 74

Git-based cache saves 50% on token usage

syumei — Mon, 27 Apr 2026 17:36:20 +0000

Article URL: https://old.reddit.com/r/vibecoding/comments/1sx4agk/gitbased_cache_saves_50_on_token_usage/

Comments URL: https://news.ycombinator.com/item?id=47924676

Points: 34

# Comments: 1

Hacking MCP Servers in AI Systems – The Rug Pull: Tool Changes After Approval

syumei — Thu, 16 Apr 2026 02:44:58 +0000

Article URL: https://medium.com/@Koukyosyumei/hacking-mcp-servers-in-ai-systems-the-rug-pull-tool-changes-after-approval-b4f1841da410

Comments URL: https://news.ycombinator.com/item?id=47788046

Points: 2

# Comments: 0

New comment by syumei in "[dead]"

syumei — Tue, 31 Mar 2026 14:07:32 +0000

I've been tracking real-world AI agent failures and incidents recently.

Things like:

prompt injection → goal hijacking

agents going rogue due to misalignment

unintended/unsafe tool use

It feels like we're starting to see repeatable patterns, not just isolated bugs.

I'm collecting cases + papers here:

https://github.com/h5i-dev/awesome-ai-agent-incidents

If you've seen interesting incidents, weird failures, or relevant research, I would love to add them.

A curated corpus of incidents and attack vectors for autonomous AI agents

syumei — Mon, 30 Mar 2026 20:24:05 +0000

Article URL: https://github.com/h5i-dev/awesome-ai-agent-incidents

Comments URL: https://news.ycombinator.com/item?id=47579280

Points: 1

# Comments: 0

ZkFuzz Foundation and Framework for Effective Fuzzing of Zero-Knowledge Circuits

syumei — Mon, 24 Nov 2025 00:30:02 +0000

Article URL: https://www.computer.org/csdl/proceedings-article/sp/2026/606500a901/2bojvL4Zswo

Comments URL: https://news.ycombinator.com/item?id=46028907

Points: 1

# Comments: 0

From College Project to 400 GitHub Stars: The Story of AIJack

syumei — Sat, 27 Sep 2025 11:54:23 +0000

Article URL: https://medium.com/@Koukyosyumei/from-college-project-to-400-github-stars-the-story-of-aijack-9b5797b206cb

Comments URL: https://news.ycombinator.com/item?id=45394977

Points: 1

# Comments: 0

From Paper to Proof: How We Can Formally Verify Blockchain Protocols with Lean4

syumei — Thu, 17 Jul 2025 13:24:12 +0000

Article URL: https://medium.com/@Koukyosyumei/from-paper-to-proof-how-we-can-formally-verify-blockchain-protocols-with-lean4-part-1-a7b3d0fc563f

Comments URL: https://news.ycombinator.com/item?id=44593135

Points: 2

# Comments: 0

New comment by syumei in "[dead]"

syumei — Mon, 12 Aug 2024 01:58:45 +0000

Hi I'm currently implementing a symbolic execution engine for EVM in Rust. Though my current implementation is just a translation of hevm, I would like to add some novel features. For example, I'm considering using A* search or multi-armed bandit to enhance scalability. Any kind of feedback, ideas, and requests are super welcome!!!

AIJack: Security and Privacy Risk Simulator for Machine Learning

syumei — Mon, 01 Jan 2024 01:38:57 +0000

Article URL: https://arxiv.org/abs/2312.17667

Comments URL: https://news.ycombinator.com/item?id=38829074

Points: 1

# Comments: 0

New comment by syumei in "[dead]"

syumei — Thu, 09 Nov 2023 03:36:31 +0000

AIJack is an easy-to-use open-source simulation tool for testing the security of your AI system against hijackers. It provides advanced security techniques like Differential Privacy, Homomorphic Encryption, K-anonymity and Federated Learning to guarantee protection for your AI. With AIJack, you can test and simulate defenses against various attacks such as Poisoning, Model Inversion, Backdoor, and Free-Rider. We support more than 30 state-of-the-art methods. For more information, check our documentation and start securing your AI today with AIJack.

New comment by syumei in "Gymbo: Gradient-Based Symbolic Execution"

syumei — Fri, 03 Nov 2023 11:36:36 +0000

Hello, Hacker News community! I am excited to introduce a new project called Gymbo, a Proof of Concept for a Gradient-based Symbolic Execution Engine. Gymbo is designed to push the boundaries of symbolic execution by leveraging recent advancements in gradient descent to tackle SMT-like formulas. This approach allows Gymbo to discover input values that satisfy each path constraint during symbolic execution.

Gymbo is entirely implemented in C++ and relies only on standard libraries.

What sets Gymbo apart from other symbolic execution tools is its simplicity and compactness in implementation. I believe that this project will help individuals better understand the core principles of symbolic execution and SMT problem-solving through gradient descent.

Gymbo: Gradient-Based Symbolic Execution

syumei — Fri, 03 Nov 2023 11:36:36 +0000

Article URL: https://github.com/Koukyosyumei/Gymbo

Comments URL: https://news.ycombinator.com/item?id=38127224

Points: 2

# Comments: 1

How should I manage and develop my open-source project?

syumei — Sun, 22 Jan 2023 15:38:40 +0000

I have developed one OSS tool (https://github.com/Koukyosyumei/AIJack), and I would like to ask how I manage it and where I should focus.

This OSS aims to simulate the security and privacy risks of Machine Learning. This tool allows users to experiment with various combinations of attacks and defenses against ML/DL models, such as Model Inversion Attacks, Poisoning Attacks, BackDoor Attacks, Differential Privacy, Encryption, K-Anonnymization, and Federated Learning. I have implemented these algorithms with Python and C++. I use black, isort, pytest, and codacy to check the quality of codes. I also generate the documentation with sphinx, but the quality is not high. The coverage of test codes is around 70%.

Since I do not have much experience with OSS projects, I don't know the best practices for OSS management and development. For example, I guess I should do the following:

- Improving the coverage of test codes

- Making documents more informative and clearer

- Writing blog posts (in many languages)

- Implementing more algorithms

- Unifying APIs for different algorithms

- Create some good-first-issuesfor new contributors

- Gathering contributors, etc

How should I prioritize them? How do you keep motivated? Some people demand a lot without gratitude or contributions (of course, most people are really lovely!). I will be a senior in April, so I must also prepare for MS or Ph.D. applications. I also have to work for my tuition. How much time do you usually spend on OSS? Should I focus on other things like research internships?

Any suggestions are really welcome —many thanks for considering my question.

Comments URL: https://news.ycombinator.com/item?id=34478365

Points: 2

# Comments: 1

AIJack: Security and Privacy Risk Simulator for Machine Learning

syumei — Mon, 02 Jan 2023 18:13:40 +0000

AIJack allows you to assess the privacy and security risks of machine learning algorithms such as Model Inversion, Poisoning Attack, Evasion Attack, Free Rider, and Backdoor Attack. AIJack also provides various defense techniques like Differential Privacy, Homomorphic Encryption, and other heuristic approaches. In addition, AIJack provides APIs for many distributed learning schemes like Federated Learning and Split Learning. You can integrate many attack and defense methods into such collaborative learning with a few lines. We currently implement more than 30 state-of-arts methods.

https://github.com/Koukyosyumei/AIJack

Comments URL: https://news.ycombinator.com/item?id=34221402

Points: 3

# Comments: 1

Let's Hijack AI! Security and Privacy Risk Simulator for Machine Learning

syumei — Mon, 24 Oct 2022 08:33:10 +0000

I have developed a framework named AIJack to simulate various attacks against machine learning models, mainly based on PyTorch and sklearn. Currently, I have implemented more than 20 algorithms! I am looking forward to your feedback!