SYNTHESIS NOTE
TopicsReward Modelsthis note

Does training order reshape how models handle different task types?

Explores whether the sequence of multi-task RL training systematically affects model capabilities across structured and creative domains, and whether this ordering effect can be predicted and optimized.

Synthesis note · 2026-02-22 · sourced from Reward Models
How should we spend compute at inference time? How do you build domain expertise into general AI models?

The standard framing of Does policy entropy collapse limit reasoning performance in RL? treats entropy collapse as a uniform phenomenon — RL training decreases entropy. Omni-Thinker (2025) reveals this is domain-dependent: structured domains (math, coding) decrease output entropy, while open-ended domains (creative writing, dialogue) increase it.

This is not a minor observation — it makes training order a mechanistic variable, not just a scheduling convenience. If you train creative writing first and structured reasoning second, the structured training will collapse the entropy that creative training expanded, potentially degrading creative capability. If you train structured reasoning first and creative writing second, the creative training preserves and expands the model's expressive range. The ordering effect is predictable from backward transfer (BWT) measurements.

Omni-Thinker uses BWT-guided scheduling: order tasks so that later tasks experience minimal negative backward transfer from earlier tasks. The approach uses hybrid rewards — verifiable (rule-based) for deterministic domains + preference-based (LLM-as-Judge) for subjective domains — enabling unified training across domain types within a single policy. The "short-form" QA tasks condition on distractors to reduce reward hacking from random guessing.

The gains are substantial: 6.2% over joint multi-task training, 12.4% over model merging. The accuracy of final multi-task models is well-predicted by forgettability rankings, even under simplifying assumptions — suggesting BWT-guided scheduling has principled theoretical grounding.

This extends Does gradually tightening token budgets beat fixed budget training? from temporal budgets to task ordering: the dimension that matters for multi-task RL is not just how much compute per task, but which tasks come first. And it enriches the entropy collapse understanding: entropy collapse is not a bug to fix everywhere — in structured domains, it reflects desirable precision. The problem is when structured-domain entropy collapse propagates to damage open-ended capabilities.

Inquiring lines that read this note 92

This note is a source for these research framings, grouped by the broader line of inquiry each explores. Scan the bold lines of inquiry; follow any specific question forward.

What are the consequences of models training on synthetic data? What capability tradeoffs emerge when scaling model reasoning abilities? What constrains reinforcement learning's ability to expand model reasoning? How do self-generated feedback mechanisms enable effective model learning? Does domain specialization cause models to lose capabilities elsewhere? What determines success in training models on multiple tasks? Why does training format shape reasoning strategy more than domain content? Why does supervised fine-tuning improve accuracy while degrading reasoning quality? How do prompt structure and constraints affect model instruction reliability? Does fine-tuning modify underlying model capabilities or only behavioral outputs? How does example difficulty affect learning efficiency in language models? How does AI adoption affect human skill development and labor equality? What pretraining choices and baseline capability constrain reinforcement learning gains? What makes weaker teacher models effective for stronger student training? How do neural networks separate factual knowledge from reasoning abilities? How does policy entropy collapse constrain reasoning-focused reinforcement learning? When does architectural design matter more than raw model capacity? Why does reinforcement learning suppress output diversity compared to supervised fine-tuning? Does reinforcement learning teach reasoning or just when to reason? How can identical external performance mask different internal representations? Can alternative training methods improve on supervised fine-tuning for language models? How do training priors constrain what context information can override? Does decoupling planning from execution improve multi-step reasoning accuracy? Do harness improvements transfer across model scales or memorize shortcuts? Why does consolidated memory sometimes degrade agent performance? Why does finetuning cause catastrophic forgetting of model capabilities? Why do continual learning scenarios trigger catastrophic forgetting and interference? Can AI systems balance emotional competence with factual reliability? Why do reward structures fail to shape long-term agent learning? How does memorization interact with learning and generalization? How should models express uncertainty rather than forced confident answers? Can single-axis benchmarks accurately predict agent deployment success? How do training data properties shape reasoning capability development? What limits mechanistic interpretability's ability to characterize models? Can language model RL training avoid reward hacking and misalignment?

Related concepts in this collection 5

This note in its neighbourhood — explore the map, then jump to a related concept in the list below.

Concept map
17 direct connections · 156 in 2-hop network ·dense cluster Open in graph ↗

Click a node to walk · click center to open · click Open in graph to see this note in the full knowledge graph

your link semantically near linked from elsewhere

Related papers in this collection 8

Papers most semantically related to this note, ranked by cosine similarity in the embedding space.

Original note title

multi-task rl reveals complementary entropy dynamics — structured domains systematically decrease output entropy while creative domains increase it making training order a mechanistic variable