Search papers, labs, and topics across Lattice.
This study investigates the internal representations of large language models (LLMs) fine-tuned for forecasting, revealing that representation-pooling probes trained on intermediate activations significantly enhance calibration compared to traditional chain-of-thought (CoT) reasoning. The findings indicate that while CoT reasoning may not faithfully reflect the underlying evidence, these probes can effectively track behavioral shifts and predict changes in forecasts, achieving an 84% accuracy rate. Additionally, the research demonstrates that forecasts are largely determined before reasoning begins, allowing for a more efficient token generation process without sacrificing accuracy.
Internal representations can serve as powerful lie detectors, revealing discrepancies in LLM forecasts that CoT reasoning may obscure.
Large language models fine-tuned for forecasting can be accurate yet poorly calibrated, and their chain-of-thought (CoT) reasoning may not faithfully reflect the evidence behind a forecast. We ask whether internal representations offer a more direct window into both. Working with Eternis-Forecaster 8B on OpenForesight, we train representation-pooling probes on intermediate activations and find they achieve substantially better calibration; a result that also holds for GLM-4.7-Flash and GLM-4.5-Air. We then assess CoT faithfulness through evidence ablation and diversionary injection: removing an influential source in the prompt often changes the model's forecast while leaving the reasoning trace untouched. The same probes function as lie detectors: their activations track behavioral shifts far better than the reasoning trace does, and they also predict the direction of change in 84% of cases, including when the CoT conceals the perturbation's influence. Finally, forced answering reveals that forecasts are largely fixed before reasoning begins: a single pre-reasoning pass recovers the committed answer and confidence, and routing questions by the spread of this pre-set answer distribution saves 30-47% of generated tokens, with no loss of accuracy. Together, these results establish probing internal representations as a practical tool for calibrating, auditing, and triaging language model forecasters and reasoning models more broadly.
