Search papers, labs, and topics across Lattice.
This paper explores the viability of Predict-Answer/Act (PA) and Reason-Predict-Reason-Answer/Act (RPRA) paradigms, where LLMs predict how a judge LLM would score their output before responding to a query. They evaluate zero-shot prediction, in-context report cards, and supervised fine-tuning to improve prediction accuracy. Results show that larger models predict generic LLM judges well zero-shot, while smaller models improve substantially via fine-tuning or report cards, achieving up to 55% and 52% improvement in prediction accuracy, respectively.
Smaller LLMs can learn to predict when they'll fail, paving the way for efficient "ask for help" systems that rival the performance of much larger models.
Large language models (LLMs) face a fundamental trade-off between computational efficiency (e.g., number of parameters) and output quality, especially when deployed on computationally limited devices such as phones or laptops. One way to address this challenge is by following the example of humans and have models ask for help when they believe they are incapable of solving a problem on their own; we can overcome this trade-off by allowing smaller models to respond to queries when they believe they can provide good responses, and deferring to larger models when they do not believe they can. To this end, in this paper, we investigate the viability of Predict-Answer/Act (PA) and Reason-Predict-Reason-Answer/Act (RPRA) paradigms where models predict -- prior to responding -- how an LLM judge would score their output. We evaluate three approaches: zero-shot prediction, prediction using an in-context report card, and supervised fine-tuning. Our results show that larger models (particularly reasoning models) perform well when predicting generic LLM judges zero-shot, while smaller models can reliably predict such judges well after being fine-tuned or provided with an in-context report card. Altogether, both approaches can substantially improve the prediction accuracy of smaller models, with report cards and fine-tuning achieving mean improvements of up to 55% and 52% across datasets, respectively. These findings suggest that models can learn to predict their own performance limitations, paving the way for more efficient and self-aware AI systems.
