Daily TEA – Doubao Dethrones DeepSeek as China’s AI Sensation
ByteDance, Dev Workflows, RLMs, AGI, EV Breakthroughs
Hello, dear TEA-mates! Here’s what you missed today in tech, AI, and science:
1.🥟 ByteDance’s Doubao Becomes China’s Top AI Chatbot
ByteDance’s Doubao AI assistant has overtaken DeepSeek with over 157 million monthly active users, making it the most popular generative AI app in China. Designed for mass-market appeal, Doubao offers chat, image, and video generation, deep integration with Douyin (the Chinese TikTok), and a warmly approachable interface. Its viral features—like social sharing and customizable agents—have driven wide adoption across all age groups, especially those less tech-savvy. ByteDance’s rapid iteration and consumer-first design pushed Doubao ahead of DeepSeek, as nearly 40% of users who left DeepSeek switched to Doubao. ByteDance now seeks to integrate Doubao across devices including smart glasses and cars. Read More
TEA For Thought: ByteDance won by making AI fun and shareable, not just smart. It’s now China’s top AI app for the masses.
2.💻 Cognition Unveils SWE-grep for Faster Coding Workflows
Cognition introduced SWE-grep and SWE-grep-mini—reinforcement learning-powered agentic models for lightning-fast, accurate retrieval of code context. These models reduce wait times for code editing from minutes to seconds and outperform current top models in both speed and precision. Fast Context, their subagent, works in tandem with smart agents to keep developers in flow, saving context budget and preventing irrelevant information pollution. With parallelism and custom tool optimizations, SWE-grep is set to dramatically streamline coding pipelines, benefitting tasks from autocomplete to codebase Q&A. Read More
TEA For Thought: It’s a major step toward seamless, intelligent dev workflows.
3.🔁 Recursive Language Models Unlock Vast Contexts for AI
MIT researchers introduced Recursive Language Models (RLMs) that allow LLMs to recursively call themselves—or other LLMs—to process enormous contexts. RLMs outperform frontier models like GPT-5 for long-form and complex queries, crucial for legal, academic, and scientific research. This approach beats context “rot,” maintains coherence over millions of tokens, and supports scalable, efficient reasoning without hitting input limits. RLMs promise a new paradigm beyond traditional summarization, with test-time recursive scaling enabling more robust AI performance. Read More
TEA For Thought: Recursive modeling enables continuous coherence over vast inputs—critical for legal research, academic summarization, and autonomous scientific discovery.
4.📖 AGI: Defining Human-Level AI with Scientific Rigor
Leading researchers have released a concrete framework for defining Artificial General Intelligence (AGI), proposing that AGI must match the cognitive versatility and proficiency of a well-educated adult. The methodology draws from Cattell-Horn-Carroll theory, breaking general intelligence into ten core domains such as reasoning, memory, perception, and more. By evaluating AI systems against established human cognitive tests, the framework reveals that models like GPT-5 show progress, but critical gaps remain—especially in long-term memory and foundational cognitive skills. This standardized approach aims to cut through hype and ambiguity, enabling clear measurement of AI advancement toward true general intelligence. Read More
TEA For Thought: When discussing anything, knowing the definition is very important—the common knowledge of what anything means is essential.
5.🔋 China Advances Solid-State EV Battery Breakthroughs
Chinese scientists have achieved major breakthroughs in all-solid-state battery technology for electric vehicles, overcoming hurdles in electrolyte conductivity, flexibility, and safety. Collaborating across research institutions, they developed iodine ion “glue,” a flexible polymer “skeleton,” and a fluorinated polyether shield—leading to batteries with enhanced storage capacity, durability, and safety under extreme conditions. The new tech promises EV ranges up to 1,000 km on a single charge, placing China ahead in the global race for next-gen battery innovation. Read More
TEA For Thought: Manufacturing and hardware in China is definitely no joke.
Prompt Tip of the Day: Give it a fake constraint —
“Explain this using only kitchen analogies”
Forces creative thinking. The weird limitation makes it find unexpected connections. Works with any random constraint (sports, movies, nature, whatever).
TEAHEE Moment
Stay sharp, stay informed. See you tomorrow!
For more updates, follow us on Twitter/X: https://x.com/the_era_arc