Paper 2512.20578
Can LLMs Predict Their Own Failures? Self-Awareness via Internal Circuits
- Published
- Dec 2025
- Research lab
- Independent
- Citations
- 10
- GitHub
- 46 stars
01 In brief
Summary
The paper introduces Gnosis, a lightweight self-awareness mechanism that enables frozen large language models (LLMs) to predict their own failures by decoding signals from internal hidden states and attention patterns during inference.
Gnosis compresses these internal traces into fixed-budget descriptors using a dual-stream architecture (hidden-state and attention circuit encoders) and predicts a correctness score with only ~5 million added parameters, independent of sequence length.
It is trained without human annotation by comparing generated answers to ground truth.
Across math reasoning (AMC12, AIME, HMMT), open-domain QA (TriviaQA), and academic knowledge (MMLU-Pro) benchmarks, and over frozen backbones from 1.7B to 20B parameters, Gnosis consistently outperforms strong internal baselines, 8B-parameter reward models (e.g., Skywork), and proprietary judges like Gemini 2.5 Pro in accuracy and calibration.
It also generalizes zero-shot to partial generations, enabling early detection of failing trajectories and compute-aware control.
A head trained on a small backbone transfers zero-shot to larger siblings, serving as a lightweight reward model.
The authors conclude that reliable correctness cues are intrinsic to the generation process and can be extracted efficiently without external supervision.
02 From the paper
Abstract
Large language models (LLMs) generate fluent and complex outputs but often fail to recognize their own mistakes and hallucinations. Existing approaches typically rely on external judges, multi-sample consistency, or text-based self-critique, which incur additional compute or correlate weakly with true correctness. We ask: can LLMs predict their own failures by inspecting internal states during inference? We introduce Gnosis, a lightweight self-awareness mechanism that enables frozen LLMs to perform intrinsic self-verification by decoding signals from hidden states and attention patterns. Gnosis passively observes internal traces, compresses them into fixed-budget descriptors, and predicts correctness with negligible inference cost, adding only ~5M parameters and operating independently of sequence length. Across math reasoning, open-domain question answering, and academic knowledge benchmarks, and over frozen backbones ranging from 1.7B to 20B parameters, Gnosis consistently outperforms strong internal baselines and large external judges in both accuracy and calibration. Moreover, it generalizes zero-shot to partial generations, enabling early detection of failing trajectories and compute-aware control. These results show that reliable correctness cues are intrinsic to generation process and can be extracted efficiently without external supervision.