Alibaba is going all in on Qwen…

Key Concepts

• Artificial Super Intelligence (ASI)
• Three-Phase Roadmap to ASI (Alibaba)
• Emergence of Intelligence
• Autonomous Action
• Self-Iteration
• Qwen 3 Max (Base Model)
• Qwen 3 Max Instruct (Reasoning Variant)
• Qwen 3 VL (Vision Language Model)
• Qwen 3 Omni (Multimodal Model)
• Mixture of Experts Architecture
• Clockbench Benchmark
• Open Source/Open Weight

Alibaba's Three-Phase Roadmap to ASI

Alibaba has announced a $52 billion, three-phase roadmap to achieve Artificial Super Intelligence (ASI) by 2032. This roadmap was unveiled at their Absara conference.

• Phase 1: Emergence of Intelligence: This phase involves AI achieving generalized understanding by learning from vast amounts of human knowledge. The video states this is where AI was a few years ago.
• Phase 2: Autonomous Action: This phase focuses on AI learning to use languages and tools to assist humans. The video claims this is where AI is currently.
• Phase 3: Self-Iteration: This final phase envisions AI connecting to the physical world and learning independently. Alibaba predicts this will be achieved by 2032. The video expresses doubt about this timeline but acknowledges Alibaba's rapid progress.

Qwen 3 Model Releases

Alibaba released several new models during the conference, with the Qwen 3 family being the most significant.

• Qwen 3 Max: The base model has over 1 trillion parameters and was pre-trained on 36 trillion tokens using a mixture of experts architecture. The video mentions that while benchmarks are often criticized, Qwen 3 Max Instruct performs well on standard benchmarks. A reasoning variant is also being trained.
• Qwen 3 VL: This is a vision language model designed to process and understand visual input like images and videos. It currently leads the Clockbench benchmark, achieving 39% accuracy in reading analog clocks. The model is open source and open weight.
• Qwen 3 Omni: This is a multimodal model capable of seeing, hearing, reading, and talking. The video jokes that this could lead to the creation of high-quality AI girlfriends.

Technical Terms and Concepts

• Artificial Super Intelligence (ASI): A hypothetical level of AI that surpasses human intelligence in all aspects.
• Parameters: The variables that an AI model learns during training. More parameters generally mean a more complex and capable model.
• Tokens: The basic units of text used for training language models.
• Mixture of Experts Architecture: A neural network architecture that combines multiple specialized "expert" networks to improve performance.
• Vision Language Model: An AI model that can process and understand both visual and textual information.
• Multimodal Model: An AI model that can process and understand multiple types of data, such as images, text, and audio.
• Open Source: Software with source code that is freely available and can be modified and distributed.
• Open Weight: AI models where the weights (parameters) of the trained model are publicly available.
• Clockbench: A benchmark specifically designed to test an AI's ability to read analog clocks.

Brilliant Sponsorship

The video includes a sponsorship segment for Brilliant, an educational platform that teaches math and computer science concepts through interactive lessons. The video recommends Brilliant's "Thinking and Code" course, which focuses on developing critical thinking skills through problem-solving. A QR code and a URL (brilliant.org/fireship) are provided for viewers to try Brilliant for free and receive 20% off a premium annual subscription.

Conclusion

Alibaba is making significant strides in AI development, with a clear roadmap towards achieving ASI. The Qwen 3 model family, particularly Qwen 3 VL and Qwen 3 Omni, demonstrates their advancements in vision language and multimodal AI. The company's commitment to open source and open weight models is also noteworthy. While the 2032 timeline for ASI may be ambitious, Alibaba's current progress suggests they are a major player in the AI race.