HAI Daily

Apple Intelligence, DeepSeek Agents and AI Video Workflows

Apple Intelligence enters China beta with Qwen, DeepSeek Creator Mode, AI industry research and video workflows, with practical tools and category guides.

By H.AI Tools Team4 stories
Products in this issue: DeepSeek, Qwen Chat, Cursor, Vidu, Excelly-AI, Crunchbase AI
In this issue

The short version

  • Apple Intelligence entered developer beta in mainland China on November 4, with Qwen built into Siri and Writing Tools. The public release is planned with iOS 27.3 in December.
  • DeepSeek's Creator Mode builds plugins from a conversation. Start with a small, inspectable utility before trusting it with real files.
  • The Hang Seng AI theme index dates from September 2023. Use it as a research framework, not as evidence that a product fits your task.
  • For AI video, settle the audience, story and shot continuity before spending generation credits.

AI news becomes useful when it helps you choose a tool or improve a workflow. This November 4 edition connects four stories to practical decisions: what Apple Intelligence's China beta means for choosing an assistant, building a small agent-powered utility, researching AI companies, and turning a short-video idea into a testable production brief. Availability and product context have been checked against the current references linked below.

01

Apple Intelligence in China: developer beta with Qwen, public release in December

The news and its practical context

Apple Intelligence reached its first testing stage in mainland China on November 4. In an October 31 announcement on its China newsroom, Apple said the features would open for testing in the latest iOS developer beta from November 4 and reach users in China with the public release of iOS 27.3 in December. Alibaba's (阿里巴巴) Qwen (通义千问) will be available inside selected experiences, including Siri, Writing Tools and Visual Intelligence, in the role ChatGPT plays elsewhere. The first wave covers established features such as Writing Tools, Live Translation and Visual Intelligence. The next-generation Siri announced this year is not included.

Eligibility is narrow. Apple lists iPhone 15 Pro, iPhone 15 Pro Max, iPhone Air and iPhone 16 or later, with Siri and device language set to English or Simplified Chinese. Until the public release ships, Apple's general support page still describes mainland China restrictions for released software. A developer beta is not the same as general availability, and the exact December date has not been announced.

For readers comparing AI assistants, the useful question is therefore broader than whether a feature exists: can you access it, supply the material your task needs, and move its output into the next step of your work? A system assistant and a separate chat app may fit different parts of that process.

Products to explore and a task to try

Start with the AI chatbots category if you want to compare standalone assistants. The Qwen Chat profile is a relevant starting point for Alibaba's assistant; DeepSeek's (深度求索) tool profile offers another route to explore. Qwen describes its service as an assistant built on the Qwen model family. Qwen's official site explains the current entry point.

For writing-specific work, browse AI writing tools. Separate the task from the interface before comparing products. Rewriting a message, extracting commitments from meeting notes, and checking a factual claim require different inputs and different standards of success.

Once the December release arrives, or today with a standalone app, try a small, repeatable brief: give each available assistant the same non-sensitive meeting note and request three action items, the person responsible for each, and any unanswered question. Check whether the output invents owners or deadlines, preserves uncertainty, and can be copied into your existing task system. This is a suggested evaluation exercise; it is not a benchmark result or a claim that these products are interchangeable with Apple Intelligence.

Get HAI Daily in your inbox

The day's AI news, what it changes, and which tools to try.

02

DeepSeek Creator Mode: turn an agent announcement into a small working tool

What the official product page establishes

DeepSeek presents Harness as an open-source environment for everyday work, coding, research and plugins. Its product page describes Creator Mode as a way to build plugins through conversation and extend the available interface, tools and skills. The examples show a utility being created and revised within a session.

The significance is the move from generating an answer to producing something that can be used repeatedly. A successful plugin needs a defined input, a useful output and predictable behaviour. A convincing demonstration alone does not establish that a generated utility will handle your own files, errors or permissions correctly.

Treat the feature as a starting point for bounded experiments. A timer or a text formatter is easier to inspect than a tool that changes many files or connects several business systems. The smaller scope makes it easier to identify whether a failure came from the instructions, the generated code or the environment in which it runs.

Choose a coding workflow around the task

Browse code generation tools to compare development environments. Keep the DeepSeek profile for product context, and explore Cursor when the task belongs in an existing codebase. Cursor's Agent documentation describes repository work, code changes, tests and shell commands. Those capabilities still require a clear definition of what the requested change should accomplish.

A useful first brief is: build a text formatter that accepts pasted notes and produces a checklist, without reading files or sending data elsewhere. Specify how empty input, duplicate items and unusually long lines should behave. Then inspect the result against those cases before adding more features. Keep a copy of the original input so the output can be checked.

For a repository task, ask for a short implementation plan, inspect the changed files, and run the checks that exercise the behaviour you requested. Our Cline, Cursor and Roo Code comparison gives additional directory context. Review current vendor documentation before choosing a plan or relying on a particular feature; this issue does not claim hands-on performance results.

03

AI industry research: use categories to separate products from market narratives

The index context

AI industry coverage often groups chip companies, model developers, cloud providers and application businesses into a single theme. That is useful for an overview, but it can obscure what a company actually sells and who uses it.

The Hang Seng Artificial Intelligence Theme Index methodology defines an index of 40 securities and covers capabilities including computing infrastructure, algorithms, model development and robotics. It records a September 2023 launch date. The index should therefore be understood as an established research framework, rather than described as a new November 4 launch.

For a tool buyer, inclusion in an industry theme does not establish product suitability. A company can be important to AI infrastructure without offering the application you need. Likewise, a useful application may be a small part of a much larger business. The next step is to translate an industry story into a product question that can be answered with evidence.

Build a research sheet that supports a decision

Use the Finance AI category to explore company-research products, including the directory entry for Crunchbase AI. If your task is organising or checking a spreadsheet, the Data Analysis category and Excelly AI profile are relevant places to compare options. These links are starting points for evaluating fit; they are not endorsements of investment outcomes or assertions about live data access.

Create a sheet with one row per company and separate columns for its product, intended customer, category, evidence URL and date checked. Keep reported facts apart from your interpretation. For a product shortlist, add the task you need to complete and the information still required before a trial.

A useful AI instruction is to identify missing fields and inconsistencies in that sheet, while leaving unsupported cells blank. Ask the assistant to explain which source supports each proposed correction. Verify calculations in the underlying spreadsheet and read the primary document behind consequential claims. This turns broad AI coverage into a structured research process without assuming that a generated summary is a reliable dataset.

Not sure which tool fits your task?Browse the H.AI Tools directory by category, then compare a shortlist.

04

Short-video attention: connect the audience story to an AI production workflow

What the audience research tells us

QuestMobile's third-quarter AI application report examines how people use AI across standalone apps, device interfaces and desktop clients. Its discussion distinguishes everyday access from deeper work delivered through connected workflows.

For creators, this is a useful reminder to keep audience measurements separate from production decisions. An app's reach, a format's viewing time and an AI tool's output quality answer different questions. A large audience does not tell you whether a generated clip has consistent characters, whether its dialogue is understandable, or whether the result serves your own viewers.

When developing a short-video concept, define the audience and the intended response first. A product explanation, a fictional scene and a social advertisement have different production needs. Each needs a brief that makes the purpose of every shot clear before you start generating footage.

Products and a practical first production brief

Explore video generation tools for the visual stage. Vidu is one directory option: its official product site lists text-to-video, image-to-video and reference-based generation. Those modes offer different ways to supply the visual context for a scene. Check the current plan, output conditions and rights that apply to your chosen workflow.

Build a short test around three shots: an opening situation, one clear action and a closing result. Write a brief for each shot with the subject, setting, camera treatment and continuity requirements. Where references are used, check that you have permission to use them. Keep the same character description and visual rules across the sequence, then inspect every generated shot before editing them together.

Use the writing stage to resolve the story before spending generation credits. A script assistant can help produce alternative hooks and a shot list; you still choose the version that matches the audience and the message. Compare candidate tools using the AI video generator guide, then verify current specifications with the provider. Start with one completed sequence and record the attempts, usable outputs and editing work required. That creates a meaningful basis for choosing a production stack.

What to try next

Choose the next tool from the task you need to finish

For everyday assistance, begin with the chatbot and writing categories. For a reusable utility, compare code generation environments and define a small acceptance test. For industry research, organise evidence before asking an assistant to interpret it. For video, establish the story and continuity requirements before generating clips.

Use the linked product profiles to investigate a shortlist, then test one representative task. The most useful result of an AI news roundup is a clearer decision about what to try next and how you will judge whether it worked.

How we put this issue together

The H.AI Tools Team checks each story against the official sources linked in the text and removes claims we could not confirm. Product links lead to H.AI Tools directory profiles. Unless an issue says otherwise, product notes describe vendor documentation, not our own benchmark tests.

Go deeper