arXiv Computation and Language

JevOut: Natural Context Can Flip Decision Models

JevOut demonstrates that natural, short additions to the context of decision models can flip their outputs from correct to incorrect, even when the correct answer remains unchanged. By optimizing context additions while keeping the source, question, choices, and gold answer fixed, the study found that 61.4% of initially correct decisions were redirected to a wrong option, with 45% receiving high confidence. Similar fragility was observed across three other decision systems on seven datasets, with flip rates between 64.9% and 73.2%.

arXiv AI
Oct 9

Type-Safe Is Not Error-Free: Typed Decision Models Follow the Option Name, Not the Definition Bound to It

The paper demonstrates that typed decision models can produce type‑correct outputs that nevertheless do not adhere to the explicit definitions of their options. By swapping only the names of options while keeping the definitions, question, and state unchanged, the authors show that decision flips can increase dramatically—up to 70.4 percentage points—when using yes/no names versus neutral controls. This effect is consistent across multiple models and decision rules, and even random strings as names yield flip rates similar to neutral controls, indicating that option‑name polarity contributes to decision instability beyond mere reassignment.

By Yu Sun, Junhao Xu, Jiajia Shi, Zijin Yang
Hugging Face Trending Papers
Jul 14

The Illusion of Robustness: Aggregate Accuracy Hides Prediction Flips under Task-Irrelevant Context

As large language models (LLMs) grow more capable, they are increasingly deployed in context-rich settings where task inputs are often accompanied by long, partially irrelevant context. In a controlled setting, we find that state-of-the-art models often appear robust to task-irrelevant context at the aggregate level: prepending it to benchmark questions causes little change in overall accuracy.

arXiv AI
Oct 16

System1-Arena: Which Decisions Can Be Left to System 1?

System1-Arena is a benchmark that evaluates which decisions can be handled by System‑1 models—those that output probabilities in a single forward pass—versus System‑2 models that generate reasoning steps. It contains 92,000 questions across 105 tasks from 268 public sources, grouped into text classification, applied decisions, evaluation and control, and agent/action, and also includes 138 closed‑loop environments such as web shopping, coding control, robotics, games, and driving. The study tests 30 models, showing that large System‑1 models can match or exceed fast closed LLMs in non‑agentic accuracy with better‑calibrated probabilities at lower latency, while reasoning LLMs are most accurate but far more expensive, and large open LLMs perform best in closed‑loop settings.

By Tianyu Chen, Xinyang Liu, Ruyi An, Zhendong Wang, Mingyuan Zhou
arXiv AI
Oct 9

TypedBench: A Benchmark for Calibration, Framing Sensitivity, and Cost in System One Decision Models

TypedBench is a new benchmark designed to evaluate system‑one decision models that output calibrated probabilities for typed answers—categorical, ordinal, or binary—using a non‑generative interface. The benchmark assesses models on policy adherence, wording robustness, probability quality (via selective prediction, ordinal proper scoring rules, and calibration error), and the cost of decisions under asymmetric cost matrices, using seven policy‑labelled generators and nine evaluation suites. Experiments on a hosted model, an open encoder, and a range of open decoders (0.8B–9B parameters) reveal that the hosted model is wording‑sensitive and underconfident, while decoders are slow and less accurate on policy questions, highlighting the need for joint evaluation across multiple dimensions.

By Rahul Sharma, Andrew B. Ducan, Ga\'etan Marceau Caron, Sebastian J. Vollmer
arXiv Computation and Language
Nov 9

Can LLMs Take Retrieved Information with a Grain of Salt?

The paper investigates how large language models (LLMs) handle the certainty of retrieved information, a critical yet underexplored aspect of retrieval-augmented generation. Eight LLMs were evaluated on their ability to adjust responses according to expressed context certainty, revealing systematic shortcomings such as failure to reproduce prior outputs after encountering uncertainty and distorted responses to expressed certainties. The authors propose an interaction strategy—combining prior reminders, context-certainty recalibration, and context simplification—that reduces obedience errors by 25% on average without altering model weights, demonstrating that interaction design can enhance LLM reliability.

By Behzad Shayegh, Mohamed Osama Ahmed, Fred Tung, Leo Feng