AI Just Learned to Write Faster — And It’s Free
This could save you 30 minutes every time you chat with AI...
Deep Dive
A new approach called TreeGraft speeds up AI language models by combining multiple drafters of different costs to build a shared draft tree. Instead of relying on one drafter for every step, TreeGraft lets a stronger drafter refine candidates, recover missed paths, and expand the tree without discarding existing branches. A lightweight scheduler decides when to use the stronger drafter. Across 10 model pairs and 6 benchmarks, TreeGraft beat the better of two fixed single-drafter strategies by 15.1% on average, with gains up to 26.6%.
Key Points
- TreeGraft uses multiple smaller AIs to draft responses at once, then lets a stronger AI refine only the best paths
- It cuts AI response time by up to 26% without needing new hardware or faster computers
- Works with today’s AI models — no extra cost or new devices required
Why It Matters
AI assistants could feel twice as fast starting today, saving you minutes every day with no new gadgets.