推荐先读Copilot Fusion goes generally available: choose effort instead of guessing models
GitHub announced Copilot Fusion on 28 October. Understand effort levels, eligible plans, model orchestration and the costs teams should measure.
Friday Works / 主题文章 / 01
AI 新闻
追踪新模型、产品发布和使用条款变化，附官方来源与实际影响。
16 篇
查看原始公告，了解实际影响
本主题文章
文章标题和正文以英文显示。进入文章页面后可切换到越南语。
推荐先读
OpenHands 1.34 adds per-conversation Docker runtime choices
The October 27 release adds Canvas Runtime settings. Distinguish Local, Worktree and Docker, check agent-server compatibility and review privileges before a pilot.
Open WebUI 0.12.4 adds on-demand skills and visible chat compaction
The October 25 release adds opt-in skill lazy loading, visible compaction summaries and more granular interface controls for managed AI workspaces.
Onyx 5.0 unifies AI plugins and changes access and upgrade requirements
Released October 24, Onyx 5.0 unifies plugins and updates token and API destination controls. Review permissions, storage migration and licensing before upgrading.
Gemini Notebook Adds Outline Review Before Rendering Slides and Videos
Announced October 19 and detailed by Workspace October 21, the co-creation flow adds Review before final output. Rollout began October 16; source and outline checks still need an editor.
Amazon Polly Upgrades Generative and Adds 14 Voices: Recheck Existing Audio Workflows
AWS’s October 20 announcement upgrades Generative and adds Hindi and Finnish. Check voice availability, region, SSML and consistency before replacing production audio.
Classroom Gemini Notebook Adds Insights and Teacher Review of Assigned Chats
An October 19 announcement adds Insights for Classroom-assigned notebooks and trusted-domain sharing. Schools should review permissions, student notices and AI summaries.
Gmail Live Expands to Workspace, With US and Language Limits
Google’s October 15 announcement expands Gmail Live on Android and iOS to eligible Workspace customers. Source cards support verification; this is not a Vietnamese rollout.
Google Updates Its Privacy Explanation for Workspace Data Connected to AI
An October 16 clarification explains how Workspace data connected to other Google services can be processed and, depending on settings, used for AI training. Account scope matters.
TIME Recognizes SynthID as AI Watermark Verification Reaches More Users
Google announced TIME’s recognition of SynthID on October 15. Watermark checks support editorial verification but do not establish whether a claim is true.
Anthropic Reports Unintended AI Actions: Review Execution Permissions
Anthropic’s October 9 report highlights risks from tool access and Internet-connected evaluations. Businesses should enforce permissions where actions execute.
AI Is Reaching More Workflows, but Most Businesses Are Still Experimenting
AI use is spreading across Vietnamese businesses. The challenge is choosing the right job, measuring the right result and knowing when to scale.
Vietnamese Businesses Are Using More AI. Measuring Value Is the Hard Part.
More businesses are trying AI, but trying tools alone does not create business value. Here is how to frame a pilot with a goal, measurement and accountable owner.
Anthropic is spending $100m to train 10,000 AI engineers: what should Vietnamese businesses learn?
Anthropic has committed $100 million to train 10,000 engineers who deploy AI in business. The important message is not certification; each person returns to solve a real project with security and measurable outcomes.
75,000 Vietnamese businesses started using AI in a year, but most are still testing
AI use is rising quickly among Vietnamese businesses, yet only a small share has integrated it into operations. The next step is not buying another tool; it is choosing the right work to measure and implement.
AI agents are reaching business systems: set the boundaries before you delegate
OpenAI's latest announcements show AI moving from answering questions to carrying out work in connected systems. Before an agent touches a CRM, inbox or internal documents, a business needs a bounded pilot, explicit permissions and human approval.