A model of errors in transformers
This addresses the issue of understanding and mitigating errors in LLMs for tasks requiring repetitive processing, offering an alternative to theories of reasoning collapse, though it is incremental in refining error analysis.
The paper tackled the problem of error rates in large language models (LLMs) on deterministic tasks like arithmetic, attributing errors to accumulated attention mechanism noise and deriving a two-parameter model that predicts accuracy with excellent empirical agreement across models like Gemini and DeepSeek.
We study the error rate of LLMs on tasks like arithmetic that require a deterministic output, and repetitive processing of tokens drawn from a small set of alternatives. We argue that incorrect predictions arise when small errors in the attention mechanism accumulate to cross a threshold, and use this insight to derive a quantitative two-parameter relationship between the accuracy and the complexity of the task. The two parameters vary with the prompt and the model; they can be interpreted in terms of an elementary noise rate, and the number of plausible erroneous tokens that can be predicted. Our analysis is inspired by an ``effective field theory'' perspective: the LLM's many raw parameters can be reorganized into just two parameters that govern the error rate. We perform extensive empirical tests, using Gemini 2.5 Flash, Gemini 2.5 Pro and DeepSeek R1, and find excellent agreement between the predicted and observed accuracy for a variety of tasks, although we also identify deviations in some cases. Our model provides an alternative to suggestions that errors made by LLMs on long repetitive tasks indicate the ``collapse of reasoning'', or an inability to express ``compositional'' functions. Finally, we show how to construct prompts to reduce the error rate.