4/1 AI Express: Apple iOS 18.4 Update / AutoGLM ‘Contemplative’ Agent / Runway Releases Gen-4

News

AI News Letter

📌 Table of Contents:

  1. 💰 OpenAI Receives Huge Funding
  2. 📱 Apple iOS 18.4 Released: AI Integration
  3. 🤖 Zhipu AI Releases “Contemplative” Agent
  4. 🤔 OpenAI: Planning to Release Open-Weight Model?
  5. ⚠️ Bill Gates: AI May Cause Mass Unemployment Within Ten Years
  6. 🎬 Runway Releases Gen-4
  7. 🎉 Google Gemini 2.5 Pro Free
  8. 💡 RAG Optimization: Document Quality Over Quantity?
  9. 🦄 AI Unicorns: Prosperity and Risk Coexist
  10. 📝 Google NotebookLM Evolves
  11. ⚙️ ChatGPT Testing New Features like “Thinking Slider”

PART 01

OpenAI Receives Huge Funding

What are the highlights?

  1. Funding Scale: OpenAI has finalized a funding intention of up to $40 billion led by SoftBank. The post-investment valuation is expected to reach $300 billion, placing it among the ranks of the world’s top private technology companies.
  2. Funding Conditions: SoftBank will undertake 75% (up to $30 billion) of the investment amount, conditional on OpenAI successfully transitioning into a for-profit entity. Institutions like Microsoft and Coatue are also participating in follow-on investments.
  3. Use of Funds: The funding will primarily be used to advance cutting-edge AI research, expand computing infrastructure (including the Stargate data center project, costing approximately $18 billion), and improve products like ChatGPT.

What can practitioners think about?

  1. Practitioners: This indicates that the AI field will receive more abundant financial support, accelerating R&D and commercialization processes; it also means intensified industry competition, with the contest for talent and computing power becoming white-hot. OpenAI’s transition may pioneer new industry models.
  2. Ordinary People: AI technology is expected to integrate more quickly into daily life, bringing more powerful tools and services; however, it may also raise concerns about technology monopolies and accessibility.

Website

OpenAI (ChatGPT): https://openai.com/

PART 02

Apple iOS 18.4 Officially Released

PART.01

What are the highlights?

  1. AI Function Updates: Apple continues to seamlessly integrate AI capabilities into the core system experience, focusing on practicality and user interaction optimization; simultaneously, when promoting AI features globally, it demonstrates its strategy for handling regulations in different regions.
    • Priority Notifications: Uses AI to filter important notifications, reducing interruptions.
    • Image Playground Adds “Sketch” Style: Provides a new AI image generation option.
  2. Other Updates: Apple News+ adds a new food section, Photos app improvements (new filters, sorting options), Control Center adds an “Ambient Music” feature (offering relaxation/focus playlists), 7 new Emojis added, Vision Pro dedicated App added.
  3. Multi-language Support: Apple Intelligence expands to more languages and regions including French, German, Spanish, Japanese, Korean, Simplified Chinese, etc. (Available in the EU; delayed in Mainland China due to compliance, currently collaborating with Alibaba).

PART.02

What can practitioners think about?

  1. Practitioners: Observe Apple’s gradual strategy for deploying AI on mobile devices and how it balances innovation with user privacy/experience; pay attention to its localization partnership models in different markets.
  2. Ordinary People: Enhanced iPhone/iPad experience, AI makes notifications smarter and creation more fun; features like ambient music focus on user digital well-being; more language users will be able to use Apple Intelligence.

Apple: https://www.apple.com

PART 03

Zhipu AI Releases AutoGLM “Contemplative” Agent

PART.01

What are the highlights?

  1. “Contemplative” Ability: Introduces self-reflection, critique, and verification mechanisms, enabling it to handle open-ended complex problems, combined with real-time web search and tool use, improving result quality and reliability.
  2. GUI Agent Feature: Can interact directly with Graphical User Interfaces (GUI) like a human (clicking the mouse, typing on the keyboard), rather than relying solely on APIs, making it more adaptable and able to collaborate with a wider range of applications.
  3. Base Model: Based on Zhipu’s full-stack large model technology (GLM-4, GLM-Z1, etc.), specifically optimized for Agent capabilities.

PART.02

What can practitioners think about?

  1. Practitioners: Represents a new direction for AI Agent development, moving from simple conversation towards autonomously completing complex tasks; GUI Agents open new paths for automating applications without APIs (like legacy systems, specific apps); the “contemplative” mechanism offers new ideas for enhancing AI reliability.
  2. Ordinary People: Future AI assistants may not just chat, but could truly help complete complex tasks like research, booking, and analysis, becoming powerful productivity tools that free up hands and minds.

PART 04

OpenAI Plans to Release Open-Weight Model

PART.01

What are the highlights?

  1. CEO Altman announced the release of the first “open-weight” language model since GPT-2, with reasoning capabilities comparable to o3-mini (which excels at STEM reasoning).
  2. “Open-weight” means publicly releasing the trained model’s weight parameters, allowing developers to customize and fine-tune, but differs from “open-source” which provides complete source code and training data.
  3. This is a significant attempt by OpenAI regarding model openness, possibly a strategic adjustment under competitive pressure from open models like Meta Llama, aiming to balance innovation promotion with risk control. It will undergo rigorous safety assessment via a “preparedness framework” before release.

PART.02

What can practitioners think about?

  1. Practitioners: Gain more flexible model customization capabilities, lowering the barrier for some AI applications; helps build a broader ecosystem around OpenAI technology; pay attention to the specific practices of its safety framework.
  2. Ordinary People: May lead to more diverse AI applications tailored to specific scenarios; however, need to be aware of the risk of open-weight models being misused.

PART 05

AI Will Trigger Mass Unemployment Within Ten Years!

PART.01

What are the highlights?

  1. Bill Gates predicted on “The Tonight Show” that within the next ten years, AI will be advanced enough to replace many professions, including doctors and teachers, because intelligence will become “free and ubiquitous.” He believes AI can solve the shortage of professionals but acknowledges that human activities (like watching sports) will still remain.
  2. Bill Gates expressed optimism about the future of AI, emphasizing that with sufficient funding, anyone can start a business in the AI field. He encourages young people to be bolder in the AI field, as they are more competitive relative to him.

PART.02

What can practitioners think about?

  1. Practitioners: A warning about the potential impact of AI on various industries, emphasizing the urgency of adaptation and transformation; Gates encourages young people to pursue AI entrepreneurship, viewing the field as full of opportunities.
  2. Ordinary People: Prompts reflection and anxiety about personal career prospects; also anticipates the potential popularization of professional knowledge (like medical care, educational tutoring) brought by AI; need to pay attention to how society will cope with potential structural unemployment.

PART 06

Runway Releases Gen-4

PART.01

What are the highlights?

Runway officially launched its latest AI video generation model, Gen-4, on April 1, 2025. It claims to be one of the highest fidelity AI video tools currently available and has been rolled out to customers.

  1. Core Innovation: “World Consistency”. Gen-4 aims to solve the challenge of consistency for characters, objects, and scenes in AI videos. It can maintain high uniformity of elements across different shots and perspectives. Users only need to provide reference images to generate characters with consistent appearance.
  2. Follows Physics: The model accurately follows instructions without additional fine-tuning and claims significant progress in simulating real-world physics (e.g., realistic motion effects).

PART.02

What can practitioners think about?

  1. Practitioners: For content creators in film, advertising, gaming, etc., Gen-4 provides more powerful tools, expected to significantly improve the usability and realism of AI videos. However, attention should be paid to copyright disputes regarding its training data sources (Runway is currently facing related lawsuits) and the potential impact on industry employment (research shows AI has already led to some job cuts).
  2. Ordinary People: May see more AI-generated video content that is extremely realistic and coherent in the future; should also consider the copyright, ethical, and social impacts of AI technology development.

Product Official Website🔗:

https://runwayml.com

PART 07

Gemini 2.5 Pro Free to Access

PART.01

What are the highlights?

  1. Free Access: Google unexpectedly announced that its experimental top-tier large language model, Gemini 2.5 Pro (which was limited to paid subscribers upon its late March release), is now available free to all Gemini users (possibly with rate limits).
  2. Excellent Performance: Gemini 2.5 Pro demonstrates outstanding performance on several authoritative benchmarks like LMArena, achieving record high scores, particularly excelling in programming, long-context processing (supporting 1M tokens), and multimodal understanding.
  3. Simultaneous Launch of Gemini Canvas: This is a brand new interactive creation space deeply integrated with Gemini 2.5 Pro. Users can write, code, generate drafts, get feedback, create personalized learning tools (like quizzes), visualize algorithms, etc., within Canvas, enabling seamless creation and collaboration with AI assistance.

PART.02

What can practitioners think about?

  1. Practitioners: Gain free access to a top-performing LLM for R&D, testing, and application development; Canvas provides content creators, developers, and educators with a powerful AI-driven productivity tool.
  2. Ordinary People: Can experience one of the most powerful current AI models without payment, usable for learning, writing, programming assistance, creative inspiration, etc.; Canvas offers an intuitive and easy-to-use platform to explore and utilize AI capabilities.

Product Official Website: https://gemini.google.com/

PART 08

RAG Systems: Not Always “More is Better”

PART.01

What are the highlights?

Research from Hebrew University (based on industry information) reveals new findings for optimizing RAG (Retrieval-Augmented Generation) systems:

  1. Document Quantity Isn’t Always Better: In a 1k token context window, 5-7 documents yield the best results; too many actually reduces relevance.
  2. Quality Over Quantity: 3 high-quality documents perform better than 9 average ones.
  3. Large Window Paradox: When the context window expands to 10k tokens, the optimal number of documents surprisingly drops to 3-5, demonstrating an information overload effect.

PART.02

What can practitioners think about?

  1. Practitioners: Provides a basis for designing more efficient and accurate AI knowledge base Q&A systems; necessitates optimizing retrieval strategies (dynamic document pools, quality assessment) and evaluation systems.
  2. Ordinary People: Future AI customer service, intelligent assistants, etc., are expected to provide more accurate answers with less interfering information.

PART 09

ChatGPT Rumored to Test “Thinking Slider” and Other Web Features

PART.01

What are the highlights?

  1. Core Focus: “Thinking Slider”. This rumored feature might allow users to adjust the “depth” or “intensity” of the model’s reasoning, letting them choose between quick responses or deep analysis based on needs. This could potentially correspond to different reasoning levels offered by the GPT-4o model.
  2. Notifications Feed: Likely for receiving updates or important information.
  3. Desktop Dictation: Allows users to input commands or content via voice.
  4. Enhanced Security Warnings: More prominent risk warnings when users click external links.
  5. “Deep Research Sharing Link”: Suggests potential addition of collaboration features, making it easier for users to share complex AI conversations or research results.
  6. Dynamic Response Style: (Based on Moonshine demo system prompt) Might adjust ChatGPT’s response style based on the user’s “memory depth” (degree of understanding context).

PART.02

What can practitioners think about?

  1. Practitioners: If the “Thinking Slider” materializes, it would provide professional users with more refined tools to adjust AI output based on specific tasks (e.g., quick drafting vs. in-depth reporting); collaboration links would greatly facilitate team use of ChatGPT for project research.
  2. Ordinary People: Enhanced sense of control over AI, allowing more flexibility in obtaining the desired type of answer; dictation feature improves input convenience; security warnings help prevent risks; overall, these potential features indicate that AI interaction interfaces are evolving towards being more user-friendly, controllable, and collaborative.

PART 10

AI Unicorn Ecosystem Revealed

PART.01

What are the highlights?

PitchBook data shows there are currently 372 AI unicorn companies globally.

  1. Industry Distribution: Computer vision, autonomous driving, and medical AI are popular tracks.
  2. Geographical Concentration: Primarily concentrated in North America (58%) and Asia (27%, with China accounting for 83% of Asia).
  3. Valuation Logic: “Technological moat” is the core factor, weighted at 47%.
  4. Risk Signals: Potential valuation bubble (Series B valuation multiples exceed those of the dot-com bubble era); high compensation for top talent fuels talent wars; generally insufficient investment in data compliance.

PART 11

Google NotebookLM Evolves Again

PART.01

What are the highlights?

According to leaked information, Google’s research and note-taking tool NotebookLM is developing significant new features:

  1. Discover Sources Mechanism: Supports intelligent federated search across 200+ academic libraries, automatically builds knowledge graphs, and introduces an “evidence strength” index to assess source credibility.
  2. UI Innovation: Plans to support multimodal boards displaying mixed text, video, and code, allow real-time collaborative annotation by multiple users, and generate intelligent summaries with adjustable levels of detail.

PART.02

What can practitioners think about?

  1. Practitioners: For researchers, students, and knowledge workers, the future NotebookLM could become a powerful tool for literature review, information organization, and team collaboration.
  2. Ordinary People: Gain smarter, more powerful tools for personal learning and information management, helpful for deeply understanding and organizing complex information.

Product Link🔗: https://notebooklm.google.com/

AI Tech News Daily Summary

What an intense day in the AI scene! 🚀 On one hand, OpenAI is pouring in money and Google is giving out freebies, making AI capabilities skyrocket while becoming more accessible. On the other hand, from Apple iPhones to Runway’s realistic videos, and Zhipu AI’s agent “thinking and working” on its own, AI is becoming ubiquitous. However, warnings like Bill Gates’ job (“rice bowl”) alert and the risks surrounding AI unicorns remind us 🤔 — this wave of intelligence is exciting and full of variables. The future is here, buckle up!

ZC Digitals

🚀 Leading corporate digital transformation, shaping the future of industry together. We specialize in creating customized digital systems integrated with AI, achieving intelligent upgrades and deep integration of business processes. Relying on a core team with top tech backgrounds from MIT, Microsoft, etc., we help you build powerful AI-driven infrastructure, enhancing efficiency, driving innovation, and achieving industry leadership.