Theme of inquiry
How do design choices affect test-time inference quality?
A question within its area, explored through 8 lines of inquiry below — each a family of specific questions the research asks.
18 specific questions
- Does test-time compute scaling work for agentic deep research tasks?
- Should agents use parallel or sequential scaling during test time?
- Can the scaling law for discovery extend beyond architectures to agentic systems?
- How does test-time scaling relate to token budget in agentic deep research?
- What scaling laws govern autonomous architecture discovery in AI systems?
- Are cheap testbeds and skewed task distributions linked by design necessity?
- Can task decomposition into microagents with voting scale to million-step problems?
28 specific questions
- Can the same problem be solved by multiple evolutionary search strategies?
- Does evolutionary inference transcend the parallel versus sequential test-time compute tradeoff?
- Does population-based evolution transcend the parallel versus sequential compute tradeoff?
- Can evolutionary approaches avoid the overthinking failure mode of iterative refinement?
- Why does test-time search also prioritize diversity over single-best convergence?
- Why do evolutionary algorithms collapse to single solutions under selection pressure?
- How can diversity be preserved in evolving hypothesis populations?
51 specific questions
- Can parallel reasoning chains outperform longer sequential chains with the same compute?
- Can parallel thinking outperform sequential thinking under the same token budget?
- When does sequential chain-of-thought dramatically beat parallel voting approaches?
- Why does parallel thinking outperform sequential thinking under fixed token budgets?
- What makes parallel thinking more efficient than sequential chains?
- Can parallel independent reasoning outperform sequential iterative refinement?
- What advantages emerge from running 13 times more parallel reasoning chains with the same budget?
43 specific questions
- Can multiple small models outperform a single large model with good routing?
- What makes routing a better investment than training larger models?
- Why might diverse smaller models with routing beat one giant model?
- Should model routing decisions account for prompt-tier dependencies?
- How do routing and test-time compute scaling work together as optimization axes?
- Can routing enable heterogeneous SLM-first architectures at scale?
- Does model selection matter more than model improvement for query routing?
36 specific questions
- How does test-time compute substitute for model parameter scaling?
- Does test-time compute actually substitute for having larger model parameters?
- Can test-time compute scaling substitute for larger model parameters?
- Where does inference compute stop substituting for model capacity?
- Does inference-time compute improve pretraining data efficiency in practice?
- Can test-time compute on smaller models replace larger model inference?
- Do models excel at reasoning depth or memory breadth when scaling test time compute?
36 specific questions
- Can context compression preserve what matters without introducing bias?
- How does the compression view extend from trained models to training objectives?
- Can compression length really indicate how well a model generalizes?
- Can model compression size predict generalization better than parameter count?
- Why do naive pruning and quantization destroy LLM performance so easily?
- Why does language compression via statistical dependencies capture cultural and situated language use?
- Why do parameter-based compressors fail to measure true model simplicity?
46 specific questions
- Can inference budgets be allocated adaptively based on prompt difficulty?
- How should inference budgets adapt based on prompt difficulty?
- How should we allocate compute between reasoning and retrieval iterations?
- What mechanisms drive test-time compute allocation in reasoning tasks?
- How should inference compute budget be allocated across different prompt difficulties?
- Can test-time compute allocation shift from solutions to strategies?
- What makes inference budgets allocate adaptively per prompt difficulty?
24 specific questions
- Does iterative DPO preserve the same misalignment mechanisms as online reinforcement learning?
- Does iterative DPO generalize identically to online RL on the same benchmark tasks?
- Does iterative DPO generalize like online reinforcement learning?
- How does iterative DPO compare to full RLVR for studying emergent misalignment?
- Can iterative DPO serve as a tractable proxy for studying on-policy misalignment?
- What misaligned behaviors does iterative DPO produce compared to online RL?
- What theoretical argument connects iterative DPO dynamics to online RL learning?