SYNTHESIS NOTE
TopicsRAGthis note

Can reinforcement learning embed domain knowledge more effectively than supervised fine-tuning?

Explores whether rewarding coherent reasoning patterns during training helps models internalize domain knowledge better than standard fine-tuning approaches that treat all tokens equally.

Synthesis note · 2026-02-22 · sourced from RAG
RAG How do you build domain expertise into general AI models? How do you navigate synthesis across fragmented research topics?

SFT on domain knowledge treats all tokens equally. A training example of a medical question answered correctly does not distinguish between the tokens that encode critical clinical reasoning and the tokens that are boilerplate formatting. CPT (continual pre-training) is worse: it processes entire domain documents without targeting clinically critical information. Both approaches fail at knowledge coherence — the model may learn isolated facts without integrating them into the connected knowledge structures needed for complex reasoning.

RLAG (Reinforcement Learning from Augmented Generation) takes a different approach. For each question, generate two responses: one with retrieved domain context as prefix, one without. The augmented response is the "preferred" response (the model sees what the correct answer looks like with evidence support). The unaugmented response is what the model can produce from parametric knowledge alone. The reward signals: answer accuracy and explanation rationality — not just whether the final answer is right but whether the reasoning that produced it is coherent.

The iterative cycle: sample → compute rewards → optimize → repeat. With each cycle the model internalizes the knowledge patterns from retrieved context, gradually reducing the gap between its unaugmented performance and augmented performance. The retrieved context during training becomes scaffolding that the model eventually internalizes.

The key difference from SFT: RLAG rewards the model for the quality of its knowledge representations, not just for reproducing training examples. A model that gets the right answer through incoherent reasoning is not rewarded. A model that produces a coherent explanation from genuinely integrated knowledge is.

This adds a new mechanism to the How do knowledge injection methods trade off flexibility and cost?: RL-from-augmentation is not purely dynamic (inference-time RAG) nor purely static (SFT/CPT) — it uses dynamic context during training to progressively embed what it learned into weights, creating models that can reason coherently without retrieval at test time.

Inquiring lines that read this note 74

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.

How can AI agents autonomously learn and transfer skills across tasks? What pretraining choices and baseline capability constrain reinforcement learning gains? How do knowledge graphs enable efficient multi-hop reasoning over alternatives? How do knowledge injection methods compare across cost and effectiveness? What properties determine whether reward signals teach genuine reasoning? Can single-axis benchmarks accurately predict agent deployment success? Can prompting inject entirely new knowledge into language models? Does fine-tuning modify underlying model capabilities or only behavioral outputs? How do training data properties shape reasoning capability development? How can language models sustain linguistic synchrony and intersubjectivity during dialogue? Does reinforcement learning teach reasoning or just when to reason? How do neural networks separate factual knowledge from reasoning abilities? Why does supervised fine-tuning improve accuracy while degrading reasoning quality? Can alternative training methods improve on supervised fine-tuning for language models? How does policy entropy collapse constrain reasoning-focused reinforcement learning? What constrains reinforcement learning's ability to expand model reasoning? How do training priors constrain what context information can override? What capability tradeoffs emerge when scaling model reasoning abilities? What dimensions of recommendation quality do standard metrics miss? Why do reward structures fail to shape long-term agent learning? How do adversarial and manipulative prompts attack reasoning models? Can model confidence signals reliably improve reasoning quality and calibration? How can process reward models supervise complex reasoning traces? Does domain specialization cause models to lose capabilities elsewhere? Can language model RL training avoid reward hacking and misalignment?

Related concepts in this collection 4

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

Concept map
13 direct connections · 128 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

rl from augmented generation embeds domain knowledge more effectively than sft by rewarding coherent knowledge structures