REVIEW 15 cited by
Effectively Controlling Reasoning Models through Thinking Intervention
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
Reasoning-enhanced large language models (LLMs) explicitly generate intermediate reasoning steps prior to generating final answers, helping the model excel in complex problem-solving. In this paper, we demonstrate that this emerging generation framework offers a unique opportunity for more fine-grained control over model behavior. We propose Thinking Intervention, a novel paradigm designed to explicitly guide the internal reasoning processes of LLMs by strategically inserting or revising specific thinking tokens. We find that the Thinking Intervention paradigm enhances the capabilities of reasoning models across a wide range of tasks, including instruction following on IFEval and Overthinking, instruction hierarchy on SEP, and safety alignment on XSTest and SorryBench. Our results demonstrate that Thinking Intervention significantly outperforms baseline prompting approaches, achieving up to 6.7% accuracy gains in instruction-following scenarios, 15.4% improvements in reasoning about instruction hierarchies, and a 40.0% increase in refusal rates for unsafe prompts using open-source DeepSeek R1 models. Overall, our work opens a promising new research avenue for controlling reasoning LLMs.
Forward citations
Cited by 15 Pith papers
-
Steering Instruction Hierarchies at Inference Time
V-Steer boosts system-prompt value spans and suppresses conflicting user spans, raising controlled instruction-hierarchy adherence from under 18% to as high as 92%.
-
Controllable LLM Reasoning via Sparse Autoencoder-Based Steering
SAE-Steering finds, via keyword-logit recall plus effectiveness ranking, sparse-autoencoder features that steer a reasoning model into a chosen reasoning strategy, beating baseline steering by ~15% on a judge-based me...
-
Learning to Reason Efficiently with Discounted Reinforcement Learning
Discounted correctness rewards in GRPO post-training shorten LLM reasoning traces by roughly 13-40% while matching pass@1 accuracy, supported by Blackwell-optimality analysis for restricted policy classes.
-
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.
-
Defending Against Prompt Injection With a Few DefensiveTokens
Five optimized token embeddings prepended to the prompt reduce prompt-injection attack success to near zero on standard benchmarks while preserving most model utility.
-
Context manipulation attacks : Web agents are susceptible to corrupted memory
Web agents with protected prompts can still be hijacked by injecting malicious steps into their stored task plans, reaching up to 63% success on privacy leaks.
-
Think Clearly: Improving Reasoning via Redundant Token Pruning
A training-free test-time method prunes low-attention reasoning tokens from the KV cache, guided by an injected end-of-thinking token, and reports accuracy gains on math competition benchmarks.
-
Overclocking LLM Reasoning: Monitoring and Controlling Thinking Path Lengths in LLMs
A fitted progress vector in LLM hidden states can monitor and, when added during decoding, shorten the thinking phase, often improving accuracy and cutting tokens.
-
Qwen Look Again: Guiding Vision-Language Reasoning Models to Re-attention Visual Information
Qwen-LookAgain copies or routes visual tokens back into the context at learned reflection steps, cutting hallucinations and raising visual QA accuracy.
-
MMATH: A Multilingual Benchmark for Mathematical Reasoning
A new multilingual math benchmark shows that reasoning models often respond in the wrong language, and English-reasoning training improves both accuracy and language consistency.
-
Efficient Long CoT Reasoning in Small Language Models
Binary cutting with on-policy validation prunes redundant chain-of-thought steps in teacher traces, letting 7B models keep most long-CoT accuracy while generating fewer tokens.
-
Not All Tokens Are What You Need In Thinking
A method that scores each chain-of-thought token by answer-conditioned perplexity and trains models on the compressed traces preserves or improves reasoning accuracy with significantly fewer tokens.
-
Two Experts Are All You Need for Steering Thinking: Reinforcing Cognitive Effort in MoE Reasoning Models Without Additional Training
Reinforcing the two experts most correlated with thinking tokens improves reasoning accuracy and efficiency in MoE large reasoning models, with gains of up to 10 points on AIME benchmarks.
-
A Comprehensive Survey on Trustworthiness in Reasoning with Large Language Models
A structured literature survey concluding that reasoning capabilities do not automatically make LLMs more trustworthy and can introduce new vulnerabilities in safety, robustness, and privacy.
-
CARE: Decoding Time Safety Alignment via Rollback and Introspection Intervention
CARE uses guard-model detection, token-buffer rollback, and self-reflective prompting to reduce harmful responses while preserving response quality.
Discussion (0). Continue with ORCID to comment.