日本語
2026-10-02 Morning edition
Morning editionAI News Daily
2026-10-02
Today's stories: extending existing AI agents and chat experiences to fit user workflows, plus the release of an open, large-scale training stack.
2026-10-02 Morning edition
Today's Highlights
FeatureAnthropic adds "mods," TypeScript functions that extend Claude Code
2026-10-01
FeatureOpenAI adds "Try on" virtual fitting for clothing and accessories in ChatGPT
2026-10-01
ResearchAi2 releases "Olmo-core 3," a trillion-parameter-scale MoE training stack, on Hugging Face
2026-10-01
2026-10-02 Morning edition
Anthropic officially announces "mods," extending how Claude Code behaves
Announced: 2026-10-01
- Small TypeScript functions can hook into events such as tool calls and permission checks
- They can add UI or replace existing features, and can be shared via a plugin directory
- Mods aren't sandboxed and have full access to the machine, so security review is essential before adoption
Why it matters: Development teams can now freely customize the CLI agent for their own workflows, which could accelerate enterprise adoption.
KEY CONTEXTBecause these extensions carry full-access permissions, review the code and permission scope before deploying them.
2026-10-02 Morning edition
OpenAI adds "Try on" virtual fitting feature to ChatGPT
Announced: 2026-10-01
- A "Try on" button on product listings lets users take or upload a selfie to generate a fitting image
- Reference photos can be managed or deleted in settings; available on both mobile and web
Why it matters: ChatGPT is pushing into retail as an entry point for search and shopping, requiring e-commerce businesses to adapt to AI-native purchase flows.
2026-10-02 Morning edition
Ai2 releases "Olmo-core 3," a trillion-parameter-scale MoE training stack
Announced: 2026-10-01
- A training stack that scales to trillion-parameter MoE model training while preserving compute efficiency
- Announced on the Hugging Face official blog and openly available
Why it matters: Open, large-scale MoE training infrastructure gives research institutions and companies the option to build proprietary models without depending on closed, major providers.
KEY CONTEXTReleased by the Allen Institute for AI (Ai2) and distributed openly via Hugging Face.
2026-10-02 Morning edition
Trend Overview
- Rather than competing purely on model performance, investment is focused on letting users extend and customize existing agents (Claude Code) and chat experiences (ChatGPT) for their own needs
- The research community keeps building infrastructure for independent development that doesn't depend on closed, major providers, such as the open training stack Olmo-core 3
- No high-confidence, multi-sourced major business, funding, or regulatory news was confirmed within this collection window, so none was included
2026-10-02 Morning edition
Summary
Three Key Takeaways
- Claude Code's new "mods" feature lets developers customize CLI agent behavior with TypeScript
- ChatGPT's new "Try on" feature extends the platform into e-commerce through virtual fitting
- Ai2 released "Olmo-core 3," a trillion-parameter-scale MoE training stack, expanding open options for independent development
What to Watch Next
- Security review and sandboxing progress for the mods feature
- Expansion of ChatGPT "Try on" usage and how e-commerce businesses respond
- Published results from actual model training using Olmo-core 3