Tag
A new Google DeepMind paper, SkillSmith, treats prefix key-value caches as an input modality, composing textual knowledge and existing weights into a fresh prefix cache for a frozen Gemma 3 4B model at inference time, improving adaptation without a target-specific training run.
Google DeepMind introduces SkillSmith, which treats model weights as an additional modality that LLMs can natively reason over, enabling instruction-steered parametric synthesis for composing skills at inference time. The approach outperforms text-only and weight-only baselines.
This paper introduces SkillSmith, an LLM augmented to reason over both prefix weights and textual knowledge, enabling instruction-steered synthesis of new parametric skills that outperform text-only and weight-only baselines.
SkillCoach introduces a self-evolving rubric framework that evaluates and enhances LLM agent skill-use by analyzing skill selection, following, composition, and reflection, providing process-level supervision beyond outcome-only metrics.