FAST01

AI Tech Weekly 2026-W23

20 min read

AI Tech Weekly 2026-W23

기간: 2026-05-28 ~ 2026-06-04

이번 주 한 줄

이번 주는 Formal Methods, LLM, Benchmarks and Evals 쪽 신호가 가장 강합니다. 단순 모델 발표보다 agent 운영, eval/검증, 개발자 도구화가 함께 움직이는지 보는 주간입니다. 특히 arXiv, OpenAI, AWS, Hugging Face 관련 움직임이 눈에 띕니다.

먼저 볼 것

1. LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks

  • 구분: arXiv cs.AI · formal-methods, llm, benchmarks-evals
  • 연결: arXiv
  • 핵심: Large Language Models (LLMs) exhibit strong informal mathematical reasoning but struggle to generate mechanically verifiable proofs in formal languages like Lean. We present LEAP, …
  • 왜 봐야 하나: LLM reasoning을 기계적으로 검증 가능한 증명/형식 시스템과 연결하려는 흐름입니다. 아직 연구 단계이므로 실제 제품 영향은 재현성, 코드 공개, 후속 인용을 보고 판단해야 합니다. 특히 arXiv 관련 신호라 생태계 변화와 함께 봐야 합니다.
  • 다음 확인: Lean 같은 형식언어에서 실제 proof success rate와 agent loop 설계를 확인해야 합니다. 코드/데이터셋/후속 구현 공개 여부가 핵심입니다.
  • 출처: https://arxiv.org/abs/2606.03303

2. Toward Pre-Deployment Assurance for Enterprise AI Agents: Ontology-Grounded Simulation and Trust Certification

  • 구분: arXiv cs.AI · llm, benchmarks-evals, ai-agents
  • 연결: arXiv
  • 핵심: Pre-deployment verification of enterprise artificial intelligence (AI) agents remains a critical gap between large language model (LLM) capability benchmarking and production deplo…
  • 왜 봐야 하나: 모델 성능 경쟁이 reasoning, 긴 컨텍스트, 비용 효율, 도메인 특화 사용성으로 확장되고 있습니다. 아직 연구 단계이므로 실제 제품 영향은 재현성, 코드 공개, 후속 인용을 보고 판단해야 합니다. 특히 arXiv 관련 신호라 생태계 변화와 함께 봐야 합니다.
  • 다음 확인: 모델 발표 자체보다 eval 결과, 비용 구조, API/제품 통합 방식을 함께 추적해야 합니다. 코드/데이터셋/후속 구현 공개 여부가 핵심입니다.
  • 출처: https://arxiv.org/abs/2606.04037

3. SMAC-Talk: A Natural Language Extension of the StarCraft Multi-Agent Challenge for Large Language Models

  • 구분: arXiv cs.AI · llm, benchmarks-evals, ai-agents
  • 연결: arXiv
  • 핵심: As LLMs become more widely deployed, they are increasingly expected to work alongside other AI agents rather than operating in isolation. Effective coordination in these settings r…
  • 왜 봐야 하나: 모델 성능 경쟁이 reasoning, 긴 컨텍스트, 비용 효율, 도메인 특화 사용성으로 확장되고 있습니다. 아직 연구 단계이므로 실제 제품 영향은 재현성, 코드 공개, 후속 인용을 보고 판단해야 합니다. 특히 arXiv 관련 신호라 생태계 변화와 함께 봐야 합니다.
  • 다음 확인: 모델 발표 자체보다 eval 결과, 비용 구조, API/제품 통합 방식을 함께 추적해야 합니다. 코드/데이터셋/후속 구현 공개 여부가 핵심입니다.
  • 출처: https://arxiv.org/abs/2606.04202

4. OpenAI frontier models and Codex are now available on AWS

  • 구분: OpenAI News · benchmarks-evals, developer-tools
  • 연결: OpenAI, AWS
  • 핵심: OpenAI frontier models and Codex are now generally available on AWS, giving enterprises a new path to build with OpenAI through the AWS environments, controls, and procurement work…
  • 왜 봐야 하나: AI 도입이 늘수록 모델 능력보다 신뢰성 측정과 검증 체계가 병목이 됩니다. frontier model과 배포 채널 변화는 기업 도입 경로와 비용 구조에 직접 연결됩니다. 특히 OpenAI, AWS 관련 신호라 생태계 변화와 함께 봐야 합니다.
  • 다음 확인: 벤치마크가 실제 업무 환경을 얼마나 반영하는지, contamination과 분포 변화에 취약한지 확인해야 합니다. OpenAI, AWS entity page에 누적합니다.
  • 출처: https://openai.com/index/openai-frontier-models-and-codex-are-now-available-on-aws

5. Welcome NVIDIA Cosmos 3: The First Open Omni-model for Physical AI Reasoning and Action

  • 구분: Hugging Face Blog · llm, inference-infra
  • 연결: Hugging Face, NVIDIA
  • 핵심: 모델 성능 경쟁이 reasoning, 긴 컨텍스트, 비용 효율, 도메인 특화 사용성으로 확장되고 있습니다.
  • 왜 봐야 하나: 모델 성능 경쟁이 reasoning, 긴 컨텍스트, 비용 효율, 도메인 특화 사용성으로 확장되고 있습니다. 오픈 생태계에서 재사용 가능한 모델/프레임워크로 확산될 가능성이 있습니다. 특히 Hugging Face, NVIDIA 관련 신호라 생태계 변화와 함께 봐야 합니다.
  • 다음 확인: 모델 발표 자체보다 eval 결과, 비용 구조, API/제품 통합 방식을 함께 추적해야 합니다. Hugging Face, NVIDIA entity page에 누적합니다.
  • 출처: https://huggingface.co/blog/nvidia/cosmos-3-for-physical-ai

Trend Radar

태그지금 의미대표 신호다음 확인
[[wiki/concepts/Formal-Methodsformal-methods]]LLM reasoning을 기계적으로 검증 가능한 증명/형식 시스템과 연결하려는 흐름입니다.LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks
[[wiki/concepts/LLMllm]]모델 성능 경쟁이 reasoning, 긴 컨텍스트, 비용 효율, 도메인 특화 사용성으로 확장되고 있습니다.LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks
[[wiki/concepts/Benchmarks-and-Evalsbenchmarks-evals]]AI 도입이 늘수록 모델 능력보다 신뢰성 측정과 검증 체계가 병목이 됩니다.LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks
[[wiki/concepts/AI-Agentsai-agents]]AI가 단순 답변을 넘어 화면 조작, 도구 호출, 업무 흐름 실행으로 이동하는 흐름입니다.LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks
[[wiki/concepts/AI-Safetyai-safety]]agent와 frontier model이 외부 시스템에 연결될수록 안전성은 연구 주제가 아니라 운영 요구사항이 됩니다.Toward Pre-Deployment Assurance for Enterprise AI Agents: Ontology-Grounded Simulation and Trust Certification
[[wiki/concepts/Healthcare-AIhealthcare-ai]]의료 도메인은 agent 신뢰성, 책임 소재, UI 자동화 검증을 가장 엄격하게 요구하는 분야입니다.Toward Pre-Deployment Assurance for Enterprise AI Agents: Ontology-Grounded Simulation and Trust Certification
[[wiki/concepts/Inference-Infrastructureinference-infra]]모델 품질만큼 중요한 축이 실제 서빙 비용, 지연 시간, 하드웨어 효율입니다.SMAC-Talk: A Natural Language Extension of the StarCraft Multi-Agent Challenge for Large Language Models
[[wiki/concepts/Developer-Toolsdeveloper-tools]]AI coding이 IDE 보조를 넘어 SDK, marketplace agent, cloud procurement까지 제품화되고 있습니다.OpenAI frontier models and Codex are now available on AWS
[[wiki/concepts/Multimodal-AImultimodal-ai]]텍스트 중심 AI가 이미지, 영상, 오디오, 행동까지 포함하는 world model/assistant 형태로 확장되고 있습니다.5 ways Google Search can level up your thrift and vintage shopping

그래프 링크

태그별 대표 항목

formal-methods · Formal Methods

  • LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks
    • Large Language Models (LLMs) exhibit strong informal mathematical reasoning but struggle to generate mechanically verifiable proofs in formal languages like Lean. We present LEAP, …
  • Toward Pre-Deployment Assurance for Enterprise AI Agents: Ontology-Grounded Simulation and Trust Certification
    • Pre-deployment verification of enterprise artificial intelligence (AI) agents remains a critical gap between large language model (LLM) capability benchmarking and production deplo…

llm · LLM

  • LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks
    • Large Language Models (LLMs) exhibit strong informal mathematical reasoning but struggle to generate mechanically verifiable proofs in formal languages like Lean. We present LEAP, …
  • Toward Pre-Deployment Assurance for Enterprise AI Agents: Ontology-Grounded Simulation and Trust Certification
    • Pre-deployment verification of enterprise artificial intelligence (AI) agents remains a critical gap between large language model (LLM) capability benchmarking and production deplo…
  • SMAC-Talk: A Natural Language Extension of the StarCraft Multi-Agent Challenge for Large Language Models
    • As LLMs become more widely deployed, they are increasingly expected to work alongside other AI agents rather than operating in isolation. Effective coordination in these settings r…

benchmarks-evals · Benchmarks and Evals

  • LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks
    • Large Language Models (LLMs) exhibit strong informal mathematical reasoning but struggle to generate mechanically verifiable proofs in formal languages like Lean. We present LEAP, …
  • Toward Pre-Deployment Assurance for Enterprise AI Agents: Ontology-Grounded Simulation and Trust Certification
    • Pre-deployment verification of enterprise artificial intelligence (AI) agents remains a critical gap between large language model (LLM) capability benchmarking and production deplo…
  • SMAC-Talk: A Natural Language Extension of the StarCraft Multi-Agent Challenge for Large Language Models
    • As LLMs become more widely deployed, they are increasingly expected to work alongside other AI agents rather than operating in isolation. Effective coordination in these settings r…

ai-agents · AI Agents

  • LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks
    • Large Language Models (LLMs) exhibit strong informal mathematical reasoning but struggle to generate mechanically verifiable proofs in formal languages like Lean. We present LEAP, …
  • Toward Pre-Deployment Assurance for Enterprise AI Agents: Ontology-Grounded Simulation and Trust Certification
    • Pre-deployment verification of enterprise artificial intelligence (AI) agents remains a critical gap between large language model (LLM) capability benchmarking and production deplo…
  • SMAC-Talk: A Natural Language Extension of the StarCraft Multi-Agent Challenge for Large Language Models
    • As LLMs become more widely deployed, they are increasingly expected to work alongside other AI agents rather than operating in isolation. Effective coordination in these settings r…

ai-safety · AI Safety

  • Toward Pre-Deployment Assurance for Enterprise AI Agents: Ontology-Grounded Simulation and Trust Certification
    • Pre-deployment verification of enterprise artificial intelligence (AI) agents remains a critical gap between large language model (LLM) capability benchmarking and production deplo…
  • A blueprint for democratic governance of frontier AI
    • OpenAI outlines a blueprint for U.S. governance of frontier AI, proposing a federal framework for safety, resilience, and national security.

healthcare-ai · Healthcare AI

  • Toward Pre-Deployment Assurance for Enterprise AI Agents: Ontology-Grounded Simulation and Trust Certification
    • Pre-deployment verification of enterprise artificial intelligence (AI) agents remains a critical gap between large language model (LLM) capability benchmarking and production deplo…

inference-infra · Inference Infrastructure

  • SMAC-Talk: A Natural Language Extension of the StarCraft Multi-Agent Challenge for Large Language Models
    • As LLMs become more widely deployed, they are increasingly expected to work alongside other AI agents rather than operating in isolation. Effective coordination in these settings r…
  • Welcome NVIDIA Cosmos 3: The First Open Omni-model for Physical AI Reasoning and Action
    • 모델 성능 경쟁이 reasoning, 긴 컨텍스트, 비용 효율, 도메인 특화 사용성으로 확장되고 있습니다.

developer-tools · Developer Tools

  • OpenAI frontier models and Codex are now available on AWS
    • OpenAI frontier models and Codex are now generally available on AWS, giving enterprises a new path to build with OpenAI through the AWS environments, controls, and procurement work…
  • GPT-4.1 deprecated
    • We have deprecated GPT-4.1 across all GitHub Copilot experiences (including Copilot Chat, inline edits, ask and agent modes, and code completions), June 1, 2026. Model Deprecation …
  • Copilot SDK is now generally available
    • The GitHub Copilot SDK is now generally available. You can embed GitHub Copilot’s agentic engine into your own applications, services, and developer tools with a stable API and pro…

multimodal-ai · Multimodal AI

  • 5 ways Google Search can level up your thrift and vintage shopping
    • Uncover second-hand scores with AI tools in Google Search and Shopping.
  • How we used Gemini to build Google I/O 2026
    • Learn how Googlers used AI to produce Google I/O 2026.
  • Take our I/O 2026 quiz, vibe coded in Google AI Studio.
    • We used Google AI Studio to vibe code a quiz about our top I/O 2026 announcements.

전체 수집 항목

  • LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks (arXiv cs.AI, 2026-06-04)
    • 트렌드: #formal-methods, #llm, #benchmarks-evals, #ai-agents
    • 엔티티: arXiv
    • 핵심: Large Language Models (LLMs) exhibit strong informal mathematical reasoning but struggle to generate mechanically verifiable proofs in formal languages like Lean. We present LEAP, …
    • 보기: Lean 같은 형식언어에서 실제 proof success rate와 agent loop 설계를 확인해야 합니다. 코드/데이터셋/후속 구현 공개 여부가 핵심입니다.
    • 출처: https://arxiv.org/abs/2606.03303
  • Toward Pre-Deployment Assurance for Enterprise AI Agents: Ontology-Grounded Simulation and Trust Certification (arXiv cs.AI, 2026-06-04)
  • SMAC-Talk: A Natural Language Extension of the StarCraft Multi-Agent Challenge for Large Language Models (arXiv cs.AI, 2026-06-04)
    • 트렌드: #llm, #benchmarks-evals, #ai-agents, #inference-infra
    • 엔티티: arXiv
    • 핵심: As LLMs become more widely deployed, they are increasingly expected to work alongside other AI agents rather than operating in isolation. Effective coordination in these settings r…
    • 보기: 모델 발표 자체보다 eval 결과, 비용 구조, API/제품 통합 방식을 함께 추적해야 합니다. 코드/데이터셋/후속 구현 공개 여부가 핵심입니다.
    • 출처: https://arxiv.org/abs/2606.04202
  • OpenAI frontier models and Codex are now available on AWS (OpenAI News, 2026-06-01)
  • Welcome NVIDIA Cosmos 3: The First Open Omni-model for Physical AI Reasoning and Action (Hugging Face Blog, 2026-06-01)
  • A shared playbook for trustworthy third party evaluations (OpenAI News, 2026-05-29)
  • A blueprint for democratic governance of frontier AI (OpenAI News, 2026-06-03)
    • 트렌드: #ai-safety
    • 엔티티: OpenAI
    • 핵심: OpenAI outlines a blueprint for U.S. governance of frontier AI, proposing a federal framework for safety, resilience, and national security.
    • 보기: red team, third-party eval, governance, integration-aware defense를 별도 축으로 계속 추적해야 합니다. OpenAI entity page에 누적합니다.
    • 출처: https://openai.com/index/frontier-safety-blueprint
  • Beyond LLMs: Why Scalable Enterprise AI Adoption Depends on Agent Logic (Hugging Face Blog, 2026-06-01)
  • 5 ways Google Search can level up your thrift and vintage shopping (Google AI Blog, 2026-06-03)
  • GPT-4.1 deprecated (GitHub Blog Changelog, 2026-06-02)
    • 트렌드: #ai-agents, #developer-tools
    • 엔티티: GitHub
    • 핵심: We have deprecated GPT-4.1 across all GitHub Copilot experiences (including Copilot Chat, inline edits, ask and agent modes, and code completions), June 1, 2026. Model Deprecation …
    • 보기: 실제 운영에서는 권한, 감사 로그, 실패 복구, prompt injection 방어를 같이 봐야 합니다. 실제 migration 일정과 API 문서를 확인합니다.
    • 출처: https://github.blog/changelog/2026-06-02-gpt-4-1-deprecated
  • Copilot SDK is now generally available (GitHub Blog Changelog, 2026-06-02)
  • Extend GitHub with agent apps (GitHub Blog Changelog, 2026-06-02)
  • How we used Gemini to build Google I/O 2026 (Google AI Blog, 2026-06-01)
  • Introducing Mellum2: A 12B Mixture-of-Experts Model by JetBrains (Hugging Face Blog, 2026-06-01)
    • 엔티티: Hugging Face
    • 핵심: 제목 기준으로 AI/tech 흐름에 걸린 항목입니다. 본문 확인 후 유지 여부를 판단합니다.
    • 보기: 다음 주에도 같은 주제가 반복되는지 확인합니다. Hugging Face entity page에 누적합니다.
    • 출처: https://huggingface.co/blog/JetBrains/mellum2-launch
  • Take our I/O 2026 quiz, vibe coded in Google AI Studio. (Google AI Blog, 2026-05-29)

관찰 포인트

  • 모델/에이전트 제품화, 추론 비용, 오픈소스 모델, 개발자 도구, 인프라 변화를 우선 추적.
  • 이번 PoC는 RSS/공개 feed 기반 1차 수집이므로, 중요한 발표 누락 여부는 수동 검토 필요.

다음 액션

  • 중요한 항목은 별도 entity/concept note로 분리.
  • 반복 사용 후 source registry와 키워드 점수를 조정.

그래프 뷰