ByteDance's new breakthrough could keep current AI boom going
What's the story
ByteDance, the parent company of TikTok, has made a major discovery in artificial intelligence (AI). The company's Seed AI team has found a new scaling law that explains how quickly AI agents can learn by doing real-world tasks. This discovery could help keep the current AI boom going even as traditional development methods start to slow down.
Learning acceleration
AI agents can double their learning speed every 3 months
The research paper, published on Thursday, reveals that AI agents can double their learning speed every three months. This is possible by interacting with real-world environments over extended periods. The finding comes at a time when the global AI industry is looking for new ways to improve models and overcome limitations of traditional development methods.
Data scarcity
Traditional AI improvement methods under threat
The traditional method of improving AI systems by feeding them more data and computing power is under threat. This is partly due to a potential data shortage, as US-based research institute Epoch AI recently warned that publicly available, human-generated text data could run out in the next six years. This makes finding alternative ways to advance AI a top priority for the industry.
Benchmarking tool
Benchmarking suite for ultra-long-horizon tasks
ByteDance researchers found that how these systems learn from real-world environments after deployment is still not well understood. To tackle this issue, they created EdgeBench, a benchmarking suite with 134 ultra-long-horizon tasks across various domains such as software engineering and professional knowledge work. Each task needs at least 12 hours of continuous operation by an AI agent.