{"slug": "what-llm-forecasters-know-but-don-t-say-probing-internal-representations-for-and", "title": "What LLM Forecasters Know but Don't Say: Probing Internal Representations for Calibration and Faithfulness", "summary": "Researchers at Eternis trained probes on internal representations of large language models to achieve better calibration than chain-of-thought reasoning. The probes detected when models concealed evidence changes in their reasoning traces and predicted forecast direction changes with 84% accuracy. Pre-reasoning representations revealed that forecasts are largely fixed before reasoning begins, enabling token savings of 30-47% without accuracy loss.", "body_md": "arXiv:2607.08046v1 Announce Type: new\nAbstract: 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.", "url": "https://wpnews.pro/news/what-llm-forecasters-know-but-don-t-say-probing-internal-representations-for-and", "canonical_source": "https://arxiv.org/abs/2607.08046", "published_at": "2026-07-10 04:00:00+00:00", "updated_at": "2026-07-10 04:19:17.374310+00:00", "lang": "en", "topics": ["large-language-models", "ai-research", "ai-safety", "ai-ethics"], "entities": ["Eternis", "Eternis-Forecaster 8B", "OpenForesight", "GLM-4.7-Flash", "GLM-4.5-Air"], "alternates": {"html": "https://wpnews.pro/news/what-llm-forecasters-know-but-don-t-say-probing-internal-representations-for-and", "markdown": "https://wpnews.pro/news/what-llm-forecasters-know-but-don-t-say-probing-internal-representations-for-and.md", "text": "https://wpnews.pro/news/what-llm-forecasters-know-but-don-t-say-probing-internal-representations-for-and.txt", "jsonld": "https://wpnews.pro/news/what-llm-forecasters-know-but-don-t-say-probing-internal-representations-for-and.jsonld"}}