Research map / Large language models / Agents, code and math
An Interactive Paradigm for Deep Research
Lin Ai, Victor Bursztyn, Xiang Chen, Julia Hirschberg, Saayan Mitra
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use STEER interactive pausing and live persona instead of GPT-Researcher with full persona | STEER · Persona-query benchmark | Alignment: 79.48 -> 85.70 | Improved |
| add persona information to the baseline (query only to full persona) | GPT-Researcher · Persona-query benchmark | Alignment: 66.63 -> 79.48 | Improved |
| add diversity-aware exploration (MMR) (ablation) | STEER · Persona-query benchmark | Focus: 84.29 -> 87.79 | Improved |
| add Explore bonus (ablation) | STEER · Persona-query benchmark | Focus: 85.17 -> 87.79 | Improved |
| add Explore bonus (ablation) | STEER · Persona-query benchmark | Alignment: 84.98 -> 85.82 | No clear effect |
| add InfoGain utility (ablation) | STEER · Persona-query benchmark | Alignment: 82.81 -> 85.82 | Improved |
| add InfoGain utility (ablation) | STEER · Persona-query benchmark | Depth: 90.41 -> 90.27 | No clear effect |
| use cost-benefit pause rule with base cost C0 >= 0.4 instead of LLM PauseAgent | STEER · Persona-query benchmark | number of pauses: PauseAgent exceeds Tol=3; STEER averages fewer than 3 | Improved |
| raise base pause cost C0 | STEER · Persona-query benchmark | Alignment per pause: local maxima around C0 = 0.7 | Mixed |
| use STEER instead of baselines in human pairwise comparison | STEER · User study (20 query-persona pairs) | Coverage (0-2): 0.828 -> 1.451 | Improved |
| use STEER instead of baselines in human pairwise comparison | STEER · User study (20 query-persona pairs) | Findability (0-2): 0.845 -> 1.534 | Improved |
| use STEER versus o4-mini-deep-research on depth and insight | STEER · Persona-query benchmark | Insight: 86.74 -> 83.04 | Hurt |
Closest papers by meaning
- When Is Rank-1 Steering Cheap? Geometry, Granularity, and Budgeted Search (NeurIPS 2026 · similarity 0.87)
- Personalized Steering of Large Language Models: Versatile Steering Vectors Through Bi-directional Preference Optimization (NeurIPS 2024 · similarity 0.87)
- MiroEval: Benchmarking Multimodal Deep Research Agents in Process and Outcome (NeurIPS 2026 · similarity 0.86)
- Steer2Edit: From Activation Steering to Component-Level Editing (NeurIPS 2026 · similarity 0.86)
- From Intent to Evidence: A Categorical Approach for Structural Evaluation of Deep Research Agents (NeurIPS 2026 · similarity 0.85)
- Steering When Necessary: Flexible Steering Large Language Models with Backtracking (NeurIPS 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
