- 프롬프트는 실재하지 않는다 (evaluation.club)
- Claude Code 개발자 보리스 처니: "나는 자주 틀린다" (borischerny.com)
- Laya, Mac에서 오프라인으로 실행하는 실시간 의사결정 AI (gist.github.com)
- Security auditing in the age of (good enough) AI (blog.trailofbits.com)
Trail of Bits가 AI 에이전트로 커스텀 도구·형식 모델을 먼저 만들고 코드 리뷰에 쓰는 보안 감사 방식을 소개한다. Miden zkVM 감사에서 실제 이슈를 찾은 6개월 경험을 정리했다.
- XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B (reddit.com)
- XiaomiMiMo/MiMo-V2.6-Flash-RL · Hugging Face (reddit.com)
- Lifestreams: a storage model for personal data (1996) (ee.columbia.edu)
- Deepseek training 2T and plans 8T model (reddit.com)
-
With GPT-6 Astra, Higgsfield AI makes video ad creation easier for small businesses and brings new creative tools to market faster.
-
OpenAI is working with an independent Advisory Group on Mathematics and Artificial Intelligence to guide the review and communication of emerging AI results.
- Creating a Blog in Gemini:// (brennan.day)
- Building standards for the next phase of AI (openai.com)
OpenAI outlines a path to shared global AI standards, calling for coordinated evaluation, reporting, and governance to improve safety.
- Gewell - Gemma4 inference engine (reddit.com)
- Expanding OpenAI Academy with new learning paths (openai.com)
Explore new OpenAI Academy learning paths for employees, developers, leaders, educators, and students to build and demonstrate practical AI skills.
- Clarification on the Qwen-image-2.1 license (reddit.com)
- You can use any LLM just like JEV (reddit.com)
- How V7 gives AI agents institutional memory (openai.com)
Using GPT-5.6, V7 turns scattered company files into context agents can use to complete complex, source-linked work.
- 채팅 기반 대규모 언어 모델은 심령술사 사기의 작동 방식을 재현한다 (softwarecrisis.dev)
- 어떤 이미지가 AI로 생성됐는지 구별할 수 있나요? (slop-sense.labtoagi.com)
- AI와 창작 공유지의 파괴 (chesterwisniewski.com)
- Cache-to-Cache: LLM 간 직접 의미 전달 (2025) (arxiv.org)