Harry Shum Explains Five Dimensions of AI Evolution: IDEA Unveils “Everything Grabbable” Model, GPU Renderer Breaks Foreign Monopoly

Insights on AI Evolution from Harry Shum at IDEA Research Institute

> "Opportunities come not just from technology itself." — Harry Shum

After steering the IDEA Research Institute (Guangdong–Hong Kong–Macao Greater Bay Area Digital Economy Research Institute) into its fifth year, Harry Shum introduced a five-dimensional framework for understanding the evolution of intelligence:

The Five Dimensions

• Algorithm Paradigm
• Intelligent Carrier
• Interaction Paradigm
• Computing Architecture
• Data

As the founding chairman, Shum's goal is not to provide a fixed roadmap, but to empower innovators to identify intersections between technology, product, and business within AI’s progression.

---

1. Algorithm Paradigm

• Past: Supervised learning focused on representation and generation capability.
• Next phase: Reinforcement learning incorporating causality and execution.
• Future trajectory: Autonomous learning toward higher-level cognition.

---

2. Intelligent Carrier

• Shift from language and multimodal models to world models and embodied models.
• Transition in carriers: From abstract symbol spaces to engagement with the physical world.

---

3. Interaction Paradigm

Shum emphasized innovation in human–computer interaction:

Historical shifts:

• Command line
• Graphical interface
• Search
• Recommendation
• Natural interaction

Natural interaction layers:

• Passive response
• Interactive execution
• Proactive modes with suggestions

Modalities:

• Text, voice, gesture
• Potential future: brain–computer interfaces

---

4. Computing Architecture

• Trend: General-purpose → Specialized; Single → Diversified.
• GPUs initiated deep learning’s boom, but costs and energy demands rose.
• Emerging solutions:
• Specialized chips for inference
• Edge computing
• Reinforcement learning optimization

---

5. Data

Phases of data usage:

• Simulated world: Static lessons
• Exploratory world: Dynamic feedback
• Inductive world: Hypothesis verification

Evolution from:

• Human-generated data
• Synthetic data
• AI actively acquiring experimental data

---

Shum’s Key Message

> Opportunities arise from understanding how technology changes carriers, interactions, computing, and data — and how it redistributes capabilities, resources, and value.

---

Future of Work and Organizations

• AI will amplify individual capability — enabling one person to achieve end-to-end production.
• Organizations may shift from control to problem definition and collaboration integration.
• Shum joked:
• > Future CEO = Chief Entertaining Officer

---

New Platforms at IDEA

IDEA announced:

• International Advanced Technology Application Promotion Center (Shenzhen)
• Shenzhen–Hong Kong Higher Education Research Exchange Center (SHARE)

---

Innovation Showcase

Embodied Intelligence — DINO-X Grasp

Core breakthrough: Vision before action.

Leveraging prior expertise and the DINO-X model:

Capabilities:

• Accurate object recognition & grasp point prediction
• Depth-based 3D reconstruction for robotic manipulation

Advantages:

• Device-agnostic & scene-agnostic
• Real-time inference: 10 FPS on RTX 4090
• Works with Intel depth cameras and X-ARM
• Supports mobile/edge deployment

Applications:

• Port container damage inspection
• Bridge bolt detection
• Night vehicle recognition
• Collaboration with Meituan Robotics Research Institute

---

AI-Native Programming Language — MoonBit

Lead: Zhang Hongbo (creator of ReScript)

Founded end of 2022 at ChatGPT’s rise.

Unique features:

• First language with native AI assistant
• IDE designed for language model integration
• Semantic analysis core

Performance:

• 10× faster numeric computation vs. Java
• WebAssembly backend smaller than Rust (−30%), TypeScript, and Go

Growth:

• 26,000 → 100,000 users in one year
• Goal: 1M users by end of 2026
• Commercial adoption by North American cloud providers

Upcoming:

> MoonBit Agent SDK — multi-language AI agent platform — will be open-sourced next month.

---

GPU Renderer — Smaray

Leader: Wang Jiaping

Industry context:

• Movie rendering: hours/day per frame vs. games: milliseconds
• Market leaders: Arnold, RenderMan, V-Ray

Breakthrough:

• Smaray achieves real-time rendering
• Scene map clarity in seconds; interactive exploration like an FPS game
• Supports NVIDIA, AMD, domestic GPUs, Apple chips

Adoption:

• MoreVFX (The Wandering Earth)
• Light Chaser Animation (Nezha)

---

Role of Open-Source Platforms like AiToEarn

AiToEarn enables:

• AI-powered content generation & monetization
• Simultaneous publishing to Douyin, Kwai, WeChat, Bilibili, Rednote, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, and X
• Integrated AI tools, analytics, and model rankings

Impact:

• Complements innovation in MoonBit, DINO-X, and Smaray
• Streamlines workflows from creation to global monetization

---

In Summary:

The convergence of AI-native tools, embodied intelligence, rendering tech, and monetization platforms is opening new pathways in both industry and creative domains. Innovators who leverage frameworks like Shum’s five dimensions — and tap into open-source ecosystems — will be best positioned to seize these opportunities.