hypothesis

active

hypothesis:qwq-and-qwen-models-have-been-extensively-post-trained-to-excel-at-single-step-tasks-causing-degradation-in-long-multi-turn-interactions

QwQ and Qwen models have been extensively post-trained to excel at single-step tasks, causing degradation in long multi-turn interactions.

Proposed explanation for why single-turn reformulation improves performance: models' training distribution is concentrated on single-turn reasoning.

Source paper

extracted_from

SFR-DeepResearch: Towards Effective Reinforcement Learning for Autonomously Reasoning Single Agents

(2025) · Xuan-Phi Nguyen · Shrey Pandit · Revanth Gangi Reddy · Aimin Xu +3

Neighborhood — ranked by edge-count

Methods (1)

method

Agentic Inference Scaffolding
gates
The paper's inference framework that reformulates multi-turn tool-calling as single-turn contextual QA for Qwen models and implements context memory management.

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Single-turn agentic workflow yields 10% absolute improvement on FRAMES for QwQ-32B over default multi-turn template.finding0.786
Result demonstrating inference-time architectural gains from reformulating multi-turn interactions as single-turn contextual QA.
There are fewer representations competent for N tasks than M<N tasks, so training more general models should yield fewer possible solutionshypothesis0.776
Selective pressure toward convergence via task generality
Qwen 2.5 7B has higher Elo scores for 'methodical' and 'formal' traits compared to Llama 3.1 8B which prefers 'colloquial'finding0.773
Model-specific baseline personality difference revealed by revealed preferences experiment
Notably, Claude Opus 4.1 and 4—the most recently released and most capable models of those that we test—perform the best in our experiments, suggesting that introspective capabilities may emerge alongside other improvements to language models.quote0.772
Key finding about the relationship between capability and introspection.
Earlier/less capable models exhibit a larger gap between think and don't think representation strengthfinding0.768
Claude 3 models show a bigger difference than newer models like Opus 4.1.
Today's Large Language Models have become so good at playing Turing's game that it often takes experts to demonstrate the present limits of their ability to simulate human-like intelligence.claim0.764
Paper's assessment of current LLM capabilities relative to Turing Test
Qwen3.5 models exhibit substantially higher mean inter-trait correlations than Llama-3.1-8B and Gemma-3-27B (mean off-diagonal 0.215 and 0.146 vs 0.090 and 0.078), indicating greater trait entanglement.finding0.763
Architecture-specific difference in trait vector geometry
QwQ-32B on MATH-500: 21.0% reasoning token reduction at intervention strength -0.96 with only 0.34% accuracy lossfinding0.761
Demonstrates reflection redundancy in stronger model on harder math benchmark