LLMs Improving LLMs: Agentic Discovery for Test-Time Scaling

• URL: http://arxiv.org/abs/2605.08083v2
• Date: Tue, 12 May 2026 03:46:05 GMT
• Title: LLMs Improving LLMs: Agentic Discovery for Test-Time Scaling
• Abstract summary: Test-time scaling (TTS) has become an effective approach for improving large language model performance.<n>Existing TTS strategies are largely hand-crafted: researchers design patterns and tune them by intuition, leaving much of the computation-allocation space unexplored.<n>We propose an environment-driven framework, AutoTTS, that changes what researchers design: from individual TTSs to environments where TTS strategies can be discovered automatically.
• Score: 63.679448814185456
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Test-time scaling (TTS) has become an effective approach for improving large language model performance by allocating additional computation during inference. However, existing TTS strategies are largely hand-crafted: researchers manually design reasoning patterns and tune heuristics by intuition, leaving much of the computation-allocation space unexplored. We propose an environment-driven framework, AutoTTS, that changes what researchers design: from individual TTS heuristics to environments where TTS strategies can be discovered automatically. The key to AutoTTS lies in environment construction: the discovery environment must make the control space tractable and provide cheap, frequent feedback for TTS search. As a concrete instantiation, we formulate width--depth TTS as controller synthesis over pre-collected reasoning trajectories and probe signals, where controllers decide when to branch, continue, probe, prune, or stop and can be evaluated cheaply without repeated LLM calls. We further introduce beta parameterization to make the search tractable and fine-grained execution trace feedback to improve discovery efficiency by helping the agent diagnose why a TTS program fails. Experiments on mathematical reasoning benchmarks show that the discovered strategies improve the overall accuracy--cost tradeoff over strong manually designed baselines. The discovered strategies generalize to held-out benchmarks and model scales, while the entire discovery costs only $39.9 and 160 minutes. Our data, and code will be open-source at https://github.com/zhengkid/AutoTTS.

Related papers

• Test-Time Scaling for Scientific Equation Discovery [9.655883330665942]
  Test-time scaling (TTS) improves language model reasoning by allocating additional test-time compute.<n>We study TTS for automated equation discovery, an open-ended setting where models search over candidate equations and rely on observed datapoints for feedback.<n>We formulate LLM-driven equation discovery as an iterative search process that unifies Best-of-N, sequential refinement, tree search, and evolution-style methods under a common compute-allocation view.
  arXiv  Detail & Related papers  (2026-08-21T07:46:04Z)
• USE: A Unified Self-Ensembling Framework for Test-Time Prompt Tuning [54.44143638529487]
  Test-time adaptation (TTA) has emerged as a popular paradigm for improving the performance of vision-language models.<n>We propose a unified self-ensembling framework (USE) that ensures consistency between the optimization and inference stages.
  arXiv  Detail & Related papers  (2026-07-04T14:45:23Z)
• DAG-STL: A Hierarchical Framework for Zero-Shot Trajectory Planning under Signal Temporal Logic Specifications [7.389002274709231]
  We study offline STL planning under unknown dynamics using only task-agnostic trajectory data.<n>We instantiate this idea in DAG-STL, a hierarchical framework that converts long-horizon STL planning into three stages.<n> Experiments in Maze2D, OGBench AntMaze, and the Cube domain show that DAG-STL substantially outperforms direct robustness-guided diffusion.
  arXiv  Detail & Related papers  (2026-04-20T14:41:37Z)
• IoT-Brain: Grounding LLMs for Semantic-Spatial Sensor Scheduling [9.212512544511576]
  IoT-Brain is a framework to interact with the physical world with unprecedented reliability and efficiency.<n>We show that IoT-Brain boosts task success rate by 37.6% over the strongest search-intensive methods.<n>In real-world deployment, it approaches the reliability upper bound while reducing 4.1 times network bandwidth.
  arXiv  Detail & Related papers  (2026-04-09T09:38:15Z)
• Test time training enhances in-context learning of nonlinear functions [51.56484100374058]
  Test-time training (TTT) enhances model performance by explicitly updating designated parameters prior to each prediction.<n>We investigate the combination of TTT with in-context learning (ICL), where the model is given a few examples from the target distribution at inference time.
  arXiv  Detail & Related papers  (2025-09-30T03:56:44Z)
• ThinkGeo: Evaluating Tool-Augmented Agents for Remote Sensing Tasks [64.86209459039313]
  ThinkGeo is an agentic benchmark designed to evaluate tool-augmented agents on remote sensing tasks via structured tool use and multi-step planning.<n>We implement a ReAct-style interaction loop and evaluate both open and closed-source LLMs on 486 structured agentic tasks with 1,773 expert-verified reasoning steps.<n>Our analysis reveals notable disparities in tool accuracy and planning consistency across models.
  arXiv  Detail & Related papers  (2025-05-29T17:59:38Z)
• EmergentTTS-Eval: Evaluating TTS Models on Complex Prosodic, Expressiveness, and Linguistic Challenges Using Model-as-a-Judge [25.51206687438354]
  We introduce $textitEmergentTTS-Eval, a comprehensive benchmark covering six challenging TTS scenarios.<n>Our framework automates both test-case generation and evaluation, making the benchmark easily accessible.<n>We evaluate state-of-the-art open-source and proprietary TTS systems, such as 11Labs, Deepgram, and OpenAI's 4o-mini-TTS, on EmergentTTS-Eval.
  arXiv  Detail & Related papers  (2025-05-29T02:36:24Z)
• Thinking Longer, Not Larger: Enhancing Software Engineering Agents via Scaling Test-Time Compute [61.00662702026523]
  We propose a unified Test-Time Compute scaling framework that leverages increased inference-time instead of larger models.<n>Our framework incorporates two complementary strategies: internal TTC and external TTC.<n>We demonstrate our textbf32B model achieves a 46% issue resolution rate, surpassing significantly larger models such as DeepSeek R1 671B and OpenAI o1.
  arXiv  Detail & Related papers  (2025-03-31T07:31:32Z)
• SCoTT: Strategic Chain-of-Thought Tasking for Wireless-Aware Robot Navigation in Digital Twins [78.53885607559958]
  We propose SCoTT, a wireless-aware path planning framework.<n>We show that SCoTT achieves path gains within 2% of DP-WA* while consistently generating shorter trajectories.<n>We also show the practical viability of our approach by deploying SCoTT as a ROS node within Gazebo simulations.
  arXiv  Detail & Related papers  (2024-11-27T10:45:49Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.