Thinking of ACE? We Can Do It with Fewer Tokens

ALTK-Evolve and ACE are methods that enable an agent to learn from its own trajectories. The key distinction lies in how they utilize the learned information, which directly impacts token consumption. While both approaches facilitate learning, ALTK-Evolve aims to achieve similar results with fewer tokens. This efficiency is highlighted by performance improvements across various difficulty levels, such as an overall increase from 79.8 to 89.3, and significant gains in Easy, Medium, and Hard categories.