China just had a major breakthrough and it's going to seem so obvious.

@baodiep244

16.4K followers · usually 1.9K views a post

Views

7.1K

Likes

224

Comments

7

Shares

51

SoundTikTok sound · original sound - baodiep244

Promoted product

Agents-A1

Why it worked

The video highlights a significant AI development from China, focusing on efficiency and a novel approach to problem-solving (recency bias) that is easily understandable and contrasts with current industry trends.

Summary

The video introduces Agents-A1, an AI model from Shanghai AI Laboratory, highlighting its efficiency with 35 billion parameters and its ability to outperform larger models. It explains that Agents-A1 overcomes recency bias by using a knowledge action graph, making it more effective for long tasks.

Structure

1Introduction of AI breakthrough
2Introduction of Agents-A1 and its capabilities
3Explanation of how it overcomes recency bias
4Comparison to industry spending and the efficiency of Agents-A1

Product placement

Aurora genAI (acts: state-of-the-art Generative AI Model for Science; removing it would change the video) Intel (appears: logo on Aurora genAI graphic; removing it would not change the video) Agents-A1 (acts: AI model developed by Shanghai AI Laboratory; removing it would change the video) Shanghai AI Laboratory (acts: developed Agents-A1; removing it would change the video)

Original caption

Meet Agents-A1: The AI Revolution That's Smarter, Not Bigger! 🤖💡

More from @baodiep244

Call to action

Follow me if you want to know what's happening over there.

On-screen text

BREAKING: US AI INFRASTRUCTURE UNDER PRESSURE Chip Export Restrictions Reshape Global AI Development AI NEWS Aurora genAI State-of-the-art Generative AI Model for Science Trained on: Source of text Scientific texts Scientific data Code 1 Trillion Parameters Megatron & Deepspeed Foundations Potential Applications Meet Agents-A1: The AI Revolution That's Smarter, Not Bigger! 🤖💡 Jul 4, 2026

Transcript

China just had a major breakthrough and it's going to seem so obvious. What if AI models thought more? I'm Gege 10, I invest in to business in China. Follow me if you want to know what's happening over there. The Shanghai AI Laboratory just dropped agents A1 through their intern Science Hub. 35 billion parameters and it's totally open source. Haven't we seen this before? It's beating trillion parameter models on their own benchmarks. It's state of the art on six of them. Back to how they're pulling it off with such a small model. They're making it think longer. In their words, they scale the horizon, not the parameters. During long tasks, normal AI often forgets the goal. This is called recency bias. They fix this problem by training agent A1 on a knowledge action graph. Think of this like a huge cause and effect road map. It's going to map the rules to the tools. In the simplest of terms, it's not going to lose its place. The AI industry spends billions on Data center is larger than football fields. But agents A1 from Shanghai Research Lab has proven to us that