Scaling to Long Horizons — Ross Taylor & Chengxi Taylor, General Reasoning
Ross Taylor and Chengxi Taylor from General Reasoning discuss training AI agents to maintain focus and perform well over extended periods—hours rather than seconds.
They explain techniques such as value models, which help AI understand which choices lead to long-term success, and bootstrapping, which extracts useful signals from limited rewards. They demonstrate how frontier models failed when given real money to trade football matches, revealing that the environment where the AI was trained was not realistic enough. The key insight is that succeeding with long horizons requires not just larger context windows but, most importantly, better and more simulated environments for training.
Vibekollen prepared this summary with AI from the original publication. The content belongs to AI Engineer.
More from AI Engineer
Inside Arca's Long-Horizon Agents That Run the Business - Aman Singh, Arca
AI Engineer 23 Oct
Real AI Agent Attacks and What Would Have Stopped Them — Andrey Rybka, Bloomberg
AI Engineer 23 Oct
Outside of tech there has been no financial payoff to US companies for using AI, now viewed as a margin-hurting cost center, as VC-backed firms burn more tokens
Techmeme 4 h ago
Source: Replit recently moved most AI spend from Anthropic to OpenAI, which gave 30%+ off in a deal worth $100M+/year, after Anthropic discontinued its discount
Techmeme 6 h ago