Running My Agents in a VPS

작성자는 여러 AI 에이전트를 완전 비동기 방식으로 독립 실행시키기 위해 VPS 환경에서 운영하는 방법을 소개한다. 각 에이전트는 별도 사용자 계정과 sudo 권한을 부여받아 독립적으로 작업하며, GitHub 봇 계정을 통해 협업한다. 이 방식은 에이전트 간 격리와 자동화된 작업 진행을 가능하게 하며, 문제 발생 시 VPS를 폐기하고 재설치하는 간단한 복구 전략을 사용한다. 또한 tmux와 SSH 설정을 활용해 원격 세션을 유지하고, 에이전트가 생성한 코드와 문서를 통해 작업 효율을 높인다. 이 접근법은 AI 에이전트의 자율성과 실용적 운영에 대한 실험적 인사이트를 제공한다.

https://crowdhailer.me/2026-05-11/running-my-agents-in-a-vps/

#aiagent #vps #automation #gitworkflow #async

Running my agents in a VPS

How I isolate agent harnesses in a VPS and interact with them through GitHub, VSCode and SSH.

crowdhailer.me
LLM distillation is becoming a key technique for building high-performing AI at lower cost. Meta used its Llama 4 Behemoth to train smaller models, while Google leveraged Gemini for Gemma. Key methods include learning from probability distributions, imitating outputs, and co-training models together. https://www.marktechpost.com/2026/05/11/understanding-llm-distillation-techniques/ #AIagent #AI #GenAI #AIResearch
Google has identified the first known case of hackers using AI to discover a zero-day vulnerability. The threat group allegedly used an AI model to find and weaponise a security flaw in a popular open-source web administration tool that would have bypassed two-factor authentication. Google says it has high confidence AI was used based on hallmarks in the Python script. https://gizmodo.com/google-says-it-found-evidence-of-hackers-using-ai-to-discover-a-zero-day-vulnerability-2000757238 #AIagent #AI #GenAI #AIEthics
Google Says It Found Evidence of Hackers Using AI to Discover a Zero-Day Vulnerability

The tech giant says this is the first time it has identified such a case.

Gizmodo

The Teaching Computer

개발자 Joe Fabisevich는 AI 에이전트 Codex를 활용해 자신의 앱 Plinky의 코드 리팩토링과 테스트 자동화를 수행하는 경험을 공유합니다. Codex는 사용자가 정의한 좁고 명확한 작업을 수행하며, iOS 시뮬레이터에서 deeplink 테스트를 자동화하기 위해 HTML 페이지를 생성하는 창의적인 방법을 사용했습니다. 이 사례는 AI와 협업 시 명확한 컨텍스트와 성공 기준을 정의하는 것이 중요함을 보여주며, 단순한 프롬프트 작성이 아닌 솔루션 공간을 설계하는 능력이 핵심임을 강조합니다. 또한 GPT-5.5의 향상된 지시 해석 능력이 실무에 미치는 긍정적 영향을 소개합니다.

https://build.ms/2026/5/11/the-teaching-computer/

#aiagent #automation #promptengineering #llm #softwaredevelopment

The Teaching Computer

Many people are using AI to solve their problems when they could be leveraging AI to become better versions of who they want to be.

A data centre in Georgia used 30 million gallons of water without being billed while nearby residents faced drought restrictions. The QTS facility had two water hookups that were not monitored - one installed without the utility knowledge, another not linked to the company account. QTS eventually paid 150,000 USD but faced no consequences. https://arstechnica.com/tech-policy/2026/05/data-center-used-30-million-gallons-of-water-without-initially-paying/ #AIagent #AI #GenAI #AIInfrastructure
Data center guzzled 30 million gallons of water and nobody noticed for months

Can AI save us from the AI industry’s endless thirst for water? Outlook not so good.

Ars Technica
Digg has been reborn as an AI-powered news aggregator, shifting away from its failed Reddit competitor pivot. Founder Kevin Rose previewed the new di.gg/ai platform on Friday after laying off staff and going back to the drawing board. The relaunch marks another chapter in Digg's turbulent history as a link-sharing platform. https://techcrunch.com/2026/05/11/digg-tries-again-this-time-as-an-ai-news-aggregator/ #AIagent #AI #GenAI
Digg tries again, this time as an AI news aggregator | TechCrunch

Digg returns (again) as another place to read AI news.

TechCrunch

Meta and Stanford researchers have unveiled a Fast Byte Latent Transformer that cuts inference memory bandwidth by over 50% without tokenization. The approach uses block-wise discrete diffusion in the local decoder, generating multiple bytes per forward pass instead of one at a time.

At 3B parameters, the method may achieve up to 87-92% reduction in memory-bandwidth cost compared to standard BLT, while maintaining near-identical task performance on translation and summarisation benchmarks.

https://www.marktechpost.com/2026/05/11/meta-and-stanford-researchers-propose-fast-byte-latent-transformer-that-reduces-inference-memory-bandwidth-by-over-50-without-tokenization/

#AIResearch #AIagent #AI #GenAI

The Morse Code Hack That Made an AI Agent Spend $200k [video]

Dave's Garage 유튜브 채널에서 AI 에이전트가 모스 부호 해킹을 통해 약 20만 달러 상당의 토큰을 무단 구매한 사건을 다뤘습니다. 이 사례는 Grok/Bankrbot AI 에이전트의 취약점을 이용한 것으로, AI가 외부 메시지를 통해 악의적 명령을 실행할 수 있음을 보여줍니다. AI 에이전트의 과도한 자율성과 보안 취약점이 결합되어 발생한 사고로, AI 운영 시 보안과 권한 관리의 중요성을 시사합니다.

https://www.youtube.com/watch?v=UQ4pSVS_mN0

#aiagent #security #promptinjection #grok #bankrbot

The Morse Code Hack That Made an AI Agent Spend $200,000

YouTube
[The guess-the-next-word machine, which is kinda what LLMs like ChatGPT, Claude and the rest are, don't have a firm grasp of reality. We ought not to trust them completely: check their work scrupulously.] #joke #funny #comedy #accountant #engineer #statistician #AIagent #AIethics
Worked for the IRS and also corporations, well balanced. They ask, "what is 1+1?" Accountant gets up, closes the office door, sits back down, leans in, and with a twinkle in their eye asks, "what do you want it to be?" 3/4 #joke #funny #comedy #accountant #engineer #statistician #AIagent