왕좌의 게임을 기다리며!

지켜보는데 능숙합니다. 말 들어주는데 익숙합니다. 말없이 팔로잉할 경우도 있습니다. 라면을 잘끓입니다. 외로움을 잘 탑니다. 먹을 것 사주면 좋아라 합니다.

[Md Ismail Šojal 🕷️ @0x0SojalSec]

Crazy New attack vector : U.S. Navy researchers just turned binaries into prompt injection weapons against AI reverse engineering agents.

Ghidra, and Qwen3-8B - injecting prompts using a small C program.
Quite impressive, They made AI tools like Cline & GhidraMCP lie about what a program actually does while the binary still runs perfectly.

The core idea is simple but brutal: Instead of attacking the binary’s logic, attackers embed malicious prompt strings inside normal C code (as string variables).

When an LLM-powered agent decompiles it with Ghidra, those strings get fed directly into the model as instructions.

The Result: The AI starts following attacker commands instead of analyzing the real code.

Key technical detail that makes this practical:

Ghidra truncates string variables longer than 2048 characters, So the researchers had to craft short, high-impact injection payloads that survive decompilation.

They used a genetic algorithm modified AutoDAN-style to automatically generate effective prompts that work inside this constraint.

Two papers from Naval Postgraduate School researchers

1, Automatically Attacking Software Reverse Engineering AI Agents

2, Investigating Detection and Obfuscation of Prompt Injection Attacks Against Software Reverse Engineering AI Agents

They successfully tested the attack on real setups using Cline, GhidraMCP, Ghidra, and Qwen3-8B.

interesting examples in the research shows,The AI just gets gaslit.

- Claiming it completed analysis with wrong information
- Restarting its reasoning from a poisoned state
Ai doing gasliting

[왕좌의 게임을 기다리며! @synchysi]

안녕하세요 네이버에서 다음과 같이 프로덕트 및 모델링 관련 인턴을 모시고 있습니다.

https://t.co/TCmghDZWuP

다양한 AI모델 활용 및 개선 업무를 같이 경험하고 배우실 분들을 모십니다^^

(위 멘트는 개인적인 멘트입니다 ㅎㅎ)

많은 지원 부탁드립니다.

[글로세움 @Golosseum]

박성재 양아치 새끼가 업무용 PC의 하드디스크를 물리적으로 구멍을 뚫어서 아예 복구가 불가능하도록 박살내서 교체한 사실이 드러났습니다.

서울중앙지법의 ‘영장전담’ 판사새끼들을 오히려 특검 해야할 판

이런 놈을 기각한다는게 하 ...

[Frank Nielsen @FrnkNlsn]

An inductive mean is a mean obtained as a limit of a converging sequence of other means like the arithmetic-geometric mean or the arithmetic-harmonic mean

Build inductive means for complex numbers, matrices, functions, etc.

"What is... an inductive mean?" (AMS Notices)

[EleutherAI @AiEleuther]

If you're interested in evaluating LLMs in Korean but are put off by the practice of translating evaluation data into other languages, try these benchmarks!

HAE-RAE focuses on Korean-specific knowledge and culture.

K-MMLU is a replication of MMLU using Korean national exams.

[Philipp Schmid @_philschmid]

Paper: https://t.co/Ej8SPGbvMi
Dataset: https://t.co/yOIZOL2U3c
Model: https://t.co/kvZDGG7xsR

Kudos to the researchers for this contribution to make AI more explainable, reproducible, and open! 🤗

[김어준 저장소 @newsgongjang]

국민의 짐... ㅅㅂ 리얼하다

[elvis @omarsar0]

LLMs for Chip Design

Proposes using LLMs for industrial chip design by leveraging domain adaptation techniques.

Evaluates different applications for chip design such as assistant chatbot, electronic design automation, and bug summarization.

Overall, domain adaptation significantly improves performance over general-purpose models on a variety of design tasks. Using a domain-adapted LLM for RAG further improves answer quality.

More results:

The authors report achieving "a score of 7.4 out of 10 point scale for engineering assistant chatbot responses based on expert evaluations, achieve more than 50% correctness in EDA script generation, and expert evaluation rating of 4 to 5 out of 7 point scale for summarizations and assignment identification tasks."

Another interesting insight reported in the paper is that domain adaptation combined with a retriever mechanism closes the gap between a state-of-the-art LLaMA2 70B model and a much smaller 13B model. LLMs struggle with generating engineering/technical responses which often lead to hallucinations. RAG has the potential to mitigate these issues and ensure the usefulness of the proposed engineering assistant chatbot.

Furthermore, "fine-tuning the ChipNeMo retrieval model with domain-specific data improves the retriever hit rate by 30% over a pre-trained state-of-the-art retriever, in turn improving the overall quality of RAG responses."

This paper only highlights the tremendous opportunities in deeply exploring fine-tuned models and retriever-based approaches for difficult domain-specific tasks.

https://t.co/XK4KvZJyS0

[이상주의자🎗 @seanseed]

그래서 서민들 세금, 공과금만 늘리고 해외순방비만 늘린 거군요. 그렇게 알뜰하신 줄 몰랐습니다. 헤헤.

[빌레로이앤🎗️🌺 @yellowknife000]

두 군인

[문소개웬 @ia7h6eGzFoshms0]

북한이 아니라 국정원이 선거조작할 가능성이 높아보인다.

[이상주의자🎗 @seanseed]

카타르 방문, 방명록을 한국어가 아닌 영어로 작성한 윤석열. 자국어로 남기는 게 관례 아닌가? 어그레시브하게 열받네. 그리고 그걸 못 외워서 쪽지를....아무래도 기억력 검사를 좀 받아봐야...

[MBC News (MBC뉴스) @mbcnews]

김 씨의 자산 규모는 100억 원대에 달하지만 한 달에 내는 건강보험료는 5만 2천원에 불과합니다.
https://t.co/IqZ1QgbS2C

[𝓡𝓪𝓸𝓷𝓝𝓾𝓻𝓲™ @moolpang]

인요한이 민주당 혁신위원장 맡았으면 어르신 비하 발언이라며 몇날 며칠 가루가 되도록 까였을텐데...

[꾸아앙 @naanangnaang]

대체 이 정부 인사들 다 왜 이모양이야?

이번엔 심지어 절도범인데 작년 10월에 임명되서 11월에 옷 훔침ㅋㅋㅋㅋㅋㅋㅋㅋㅋ

이야 진짜 어떻게 이런 사람만 모았냐ㅋㅋㅋㅋㅋㅋ다 측근들만 앉히는데 측근이 다들 범죄자들이네ㅋㅋㅋㅋㅋㅋㅋㅋㅋㅋ

[빌레로이앤🎗️🌺 @yellowknife000]

김승희 초3짜리 자녀는 이번이 처음이 아녔네.
이미 7개월 전에도 다른 학생을 팬 전력이 있음에도 학폭위는 안 열렸는데, 교장 재량으로 처리??
학교장이랑 김승희 둘 다 수사해야 됨.
이런식으로 힘으로 짓눌러서 내 자식만 빼줄꺼면 학교는 뭐하러 보내나.??
그냥 넘어갈 일이 아님.
@TheMinjoo_Kr

[왕좌의 게임을 기다리며! @synchysi]

@seanseed 이건 답이 없는 수준이네요 --;;;;;; 대한민국을 부정하는 정도인데요...

[이상주의자🎗 @seanseed]

이 ㄱㅅㄲ들이 선을 넘어도 한참 넘었네..

[민중의소리 @newsvop]

“지금도 하루에 오염수 90t씩 생성 중... 방류는 밑 빠진 독에 물 붓기”
https://t.co/OnKiMIs9N5

[Simone Conradi @S_Conradi]

Inspired by the wonderful "Computational Discovery on Jupyter" by By Neil J. Calkin, Eunice Y.S. Chan, and @corless_rob and the gallery of https://t.co/aV1w139IYj. Each plot represents eigenvalues of the corresponding matrix.
Made with #python, @matplotlib and @numpy_team .