Paper Radar · demo

2026-11-14

Every new paper was read against your interests. These are the ones worth your time.

44papers read
→
10shortlisted
→
9must-read

31,561 input tokens · mock-heuristic via mock

Must-read ≥ 80%

95%surveyarxiv

A Survey of Uncertainty Estimation in Large Language Models

Grace Mensah, Daniel Kowalski · cs.CL

LLM calibration 95%
Abstract

We review methods for uncertainty estimation in large language models, covering verbalized confidence, sampling consistency, and probe-based approaches. We organize the literature by the kind of prediction being scored and discuss open problems in calibration for long-form generation.

arxiv:2609.99005
95%benchmarkarxiv

Rerank Less, Retrieve Better: Query-Adaptive Depth for Retrieval-Augmented Generation

Samir Haddad, Elena Petrova · cs.IR, cs.CL

RAG & retrieval 95%
Abstract

Retrieval-augmented generation pipelines typically rerank a fixed number of passages. We show that retrieval quality for question answering depends strongly on query difficulty, and train a lightweight controller that chooses retrieval depth per query. The controller cuts reranker calls by 58% while improving answer accuracy on four question answering benchmarks.

arxiv:2609.99003
95%benchmarkarxiv

TraceGrade: Evaluating LLM Agents on Long-Horizon Tool-Use Tasks with Step-Level Verification

Mira Okafor, Tomás Lindqvist, Hae-won Seo et al. · cs.CL, cs.AI

Agent evaluation 95%
Abstract

Benchmarks for LLM agents usually score only the final answer of a multi-step episode. We introduce TraceGrade, a benchmark of 1,240 multi-step tool-use tasks across web, shell, and spreadsheet environments in which every intermediate action is verified against an executable checker. Evaluating 14 LLM agents, we find that final-answer accuracy overstates step-level correctness by up to 31 points. Code and tasks are publicly released.

arxiv:2609.99001
95%new methodarxiv

Calibrated Abstention for Language Model Question Answering via Conformal Uncertainty

Jonas Brandt, Aiko Tanaka, Lucía Ferrer · cs.CL, cs.LG

LLM calibration 95%RAG & retrieval 35%
Abstract

Language model predictions are often overconfident, which makes abstention unreliable. We propose a conformal procedure that turns token-level signals into calibrated uncertainty estimation for open-domain question answering. Across five datasets, our method reduces expected calibration error by 43% and lets the language model abstain with a guaranteed error rate.

arxiv:2609.99002
94%positionarxiv

Position: Evaluating Autonomous Agents Needs Cost-Aware Metrics

Michael Brooks, Anika Sharma · cs.AI

Agent evaluation 94%
Abstract

We argue that evaluating LLM agents by success rate alone hides the cost of multi-step tool-use. We propose reporting success per dollar and per minute and show how rankings on popular agent benchmarks change under cost-aware metrics.

arxiv:2609.99014
94%benchmarkarxiv

Do LLM Agents Know When They Are Wrong? Self-Evaluation of Tool-Use Trajectories

Laura Chen, Pedro Alves, Yuki Mori · cs.AI, cs.CL

Agent evaluation 94%LLM calibration 68%
Abstract

We study whether LLM agents can evaluate their own multi-step tool-use trajectories. Agents' self-reported confidence is poorly calibrated: across three agent benchmarks, trajectories rated 90% likely to succeed succeed only 61% of the time. A simple uncertainty estimation probe on hidden states recovers much of the gap.

arxiv:2609.99009
93%benchmarkarxiv

SciRetrieve: A Benchmark of Retrieval Quality for Scientific Question Answering

Oluwaseun Adeyemi, Clara Dubois · cs.IR, cs.CL

RAG & retrieval 93%
Abstract

We release SciRetrieve, a benchmark of 9,800 expert-written scientific questions with passage-level relevance labels, to measure retrieval quality for retrieval-augmented question answering. Dense retrievers and rerankers still miss a supporting passage for 27% of questions.

arxiv:2609.99013
89%benchmarkarxiv

4-Bit KV Cache Compression for Long-Context LLM Inference

Arjun Mehta, Sofia Rossi · cs.LG, cs.CL

Efficient inference 89%
Abstract

Long-context LLM inference is dominated by KV cache memory. We introduce a 4-bit quantization scheme for the KV cache with per-channel outlier handling that keeps perplexity within 0.1 of full precision and enables 2.6x cheaper inference at 128k context.

arxiv:2609.99010
89%new methodarxiv

SpecTree: Draft Trees for Faster Speculative Decoding in Large Language Models

Wei Zhang, Olivia Grant, Rafael Moreno et al. · cs.CL, cs.LG

Efficient inference 89%LLM calibration 35%
Abstract

Speculative decoding accelerates LLM inference by verifying draft tokens in parallel. We propose SpecTree, which drafts a tree of continuations and prunes it with a learned acceptance model. SpecTree delivers 3.1x faster decoding than autoregressive sampling with identical outputs, and composes with quantization for cheaper inference. Code is available.

arxiv:2609.99004

Maybe 50% to 80%

68%new methodarxiv

Faster Decoding for Speech Language Models via Speculative Unit Prediction

Hyun-woo Park, Isabel Novak · eess.AS, cs.CL

LLM calibration 68%Efficient inference 48%RAG & retrieval 35%
Abstract

Speech language models generate discrete units slowly. We adapt speculative decoding to unit sequences and obtain 2.4x faster inference with no loss in speech quality.

arxiv:2609.99044
Near misses (8): just under your maybe bar, useful for tuning thresholds
Filtered by your exclusions (1)