REVIEW 14 cited by
START: Self-taught Reasoner with Tools
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Large reasoning models (LRMs) like OpenAI-o1 and DeepSeek-R1 have demonstrated remarkable capabilities in complex reasoning tasks through the utilization of long Chain-of-thought (CoT). However, these models often suffer from hallucinations and inefficiencies due to their reliance solely on internal reasoning processes. In this paper, we introduce START (Self-Taught Reasoner with Tools), a novel tool-integrated long CoT reasoning LLM that significantly enhances reasoning capabilities by leveraging external tools. Through code execution, START is capable of performing complex computations, self-checking, exploring diverse methods, and self-debugging, thereby addressing the limitations of LRMs. The core innovation of START lies in its self-learning framework, which comprises two key techniques: 1) Hint-infer: We demonstrate that inserting artificially designed hints (e.g., ``Wait, maybe using Python here is a good idea.'') during the inference process of a LRM effectively stimulates its ability to utilize external tools without the need for any demonstration data. Hint-infer can also serve as a simple and effective sequential test-time scaling method; 2) Hint Rejection Sampling Fine-Tuning (Hint-RFT): Hint-RFT combines Hint-infer and RFT by scoring, filtering, and modifying the reasoning trajectories with tool invocation generated by a LRM via Hint-infer, followed by fine-tuning the LRM. Through this framework, we have fine-tuned the QwQ-32B model to achieve START. On PhD-level science QA (GPQA), competition-level math benchmarks (AMC23, AIME24, AIME25), and the competition-level code benchmark (LiveCodeBench), START achieves accuracy rates of 63.6%, 95.0%, 66.7%, 47.1%, and 47.3%, respectively. It significantly outperforms the base QwQ-32B and achieves performance comparable to the state-of-the-art open-weight model R1-Distill-Qwen-32B and the proprietary model o1-Preview.
Forward citations
Cited by 14 Pith papers
-
Toward Skill-Native LLMs: Skill Entropy for Benchmarking and Training Long-Horizon Reasoning
Skill entropy, a reference-model-based measure of skill-switching difficulty, calibrates a new cross-skill benchmark and serves as an RL reward, more than doubling small models' scores.
-
PEARL: Solver-in-the-Loop Interactive Optimization Modeling from Natural Language
Training an LLM as a multi-turn agent that runs and repairs solver code raises verified optimization solve rates, with the 4B PEARL model outperforming DeepSeek-V3.2-685B in aggregate.
-
Gecko: A Simulation Environment with Stateful Feedback for Refining Agent Tool Calls
A simulated tool environment with argument validation, response synthesis, and task-state feedback improves LLM tool-call accuracy at test time.
-
Does More Inference-Time Compute Really Help Robustness?
With exposed reasoning chains, increasing inference-time compute consistently decreases measured robustness across 12 open-source reasoning models, while hidden chains show improvements.
-
MaskSearch: A Universal Pre-Training Framework to Enhance Agentic Search Capability
A pre-training task called RAMP, where models practice searching to fill masked text spans, improves downstream agentic open-domain QA performance across Qwen and LLaMA models.
-
Tool-Star: Empowering LLM-Brained Multi-Tool Reasoner via Reinforcement Learning
Tool-Star combines cold-start supervised fine-tuning with a multi-tool self-critic reinforcement learning algorithm and hierarchical rewards to improve LLM tool-use reasoning.
-
Learning to Reason via Mixture-of-Thought for Logical Reasoning
Jointly training and voting across natural language, code, and truth-table reasoning modalities improves LLM logical reasoning accuracy by up to 11.7 percentage points.
-
Logical Reasoning with Outcome Reward Models for Test-Time Scaling
Outcome reward models trained on multi-sample chain-of-thought plus deliberately flawed 'echo' rationales improve Best-of-N test-time verification for deductive reasoning.
-
Thinking Before You Speak: A Proactive Test-time Scaling Approach
TBYS inserts generated situation-and-goal insights between reasoning steps and reports math accuracy gains, but the insight library is filtered using test-set answers.
-
CRScore++: Reinforcement Learning with Verifiable Tool and AI Feedback for Code Review
A two-stage SFT+DPO pipeline using linter and code-smell tool outputs plus an LLM judge improves generated code review comments and appears to transfer from Python to Java and JavaScript, though the judge-based evalua...
-
Empowering Nanoscale Connectivity through Molecular Communication: A Case Study of Virus Infection
A position paper proposing molecular communication as the link layer for epidemic-control bio-nano networks, with an ORF3a-based mutation identification simulation; the provided manuscript body does not match this abstract.
-
Deep Research Agents: A Systematic Examination And Roadmap
A survey that organizes LLM-powered deep research agents into static versus dynamic workflows and single versus multi agent architectures, and reviews their benchmarks and open challenges.
-
CoRT: Code-integrated Reasoning within Thinking
Inserting targeted hints into a few training examples teaches reasoning models to compute with Python instead of text, improving accuracy and cutting token use by 30 to 50 percent.
-
Infinite Video Understanding
The paper argues that video understanding research should aim at processing streams of arbitrary, unbounded duration and outlines the challenges, directions, and metrics needed.
Discussion (0). Sign in to comment.