In the news
How Much Is an AI Token Worth? Scaling Laws for Wild AI-Generated Web Text
arXiv cs.AI · Published · 3 min read
In 30 seconds
- What happened
- Researchers found that 27-31% of web text is AI-generated and studied how this affects language model pretraining by training 800 models with varying AI-to-human token ratios.
- Why it matters
- Matters for engineers building language models who need to understand whether AI-generated pretraining data helps or hurts model quality on human text tasks.
- Watch out
- AI tokens initially help data-starved models but quickly become harmful as more are added. Benefits reverse entirely for well-resourced models, contradicting existing scaling laws.
- language model
- token
The patterns behind this
- Generative UI (Agent-Rendered Interfaces)
- Agent-Readable Web (llms.txt / NLWeb)
- Generative Agents Memory
Each one covers how the technique works, when it earns its cost, and where it breaks.
The Agent Architect
One pattern, one tradeoff, one production failure story. A short weekly briefing for people building agentic systems.
Weekly email, one-click unsubscribe. We only use your address to send the briefing.