Pandora's AI Model Routing Box: Efficient Allocation with Costly Value Estimation
The paper treats model routing as a paid search problem: better estimates can improve allocation, but only when they are worth buying.
Its “Pandora’s Router” uses value-of-information rules to decide when to spend on a more accurate estimate for a given specialist and query. In experiments, it matched exhaustive routing quality while calling the expensive estimator much less often. A decentralized version lets specialists choose whether to pay for self-assessment before claiming work, with efficiency gains when estimates are reliable and strategic skew when they are noisy. ArXiv · AI/CL/LG's note
Its “Pandora’s Router” uses value-of-information rules to decide when to spend on a more accurate estimate for a given specialist and query. In experiments, it matched exhaustive routing quality while calling the expensive estimator much less often. A decentralized version lets specialists choose whether to pay for self-assessment before claiming work, with efficiency gains when estimates are reliable and strategic skew when they are noisy. ArXiv · AI/CL/LG's note
score 4