Training Skills Like Parameters via Self-Supervised Semantic Diffusion

• URL: http://arxiv.org/abs/2607.27557v1
• Date: Thu, 30 Jul 2026 01:03:59 GMT
• Title: Training Skills Like Parameters via Self-Supervised Semantic Diffusion
• Abstract summary: Large Language Models (LLMs) often fall short of human experts in highly specialized, open-ended domains such as creative screenwriting.<n>We propose a novel, unsupervised self-evolving agent framework inspired by the corruption-and-reconstruction paradigm of diffusion models.<n>We evaluate our framework on the challenging task of short drama screenwriting.
• Score: 15.133304131514713
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: While Large Language Models (LLMs) demonstrate remarkable general instruction-following capabilities, they often fall short of human experts in highly specialized, open-ended domains such as creative screenwriting. Prior approaches typically adopt post-training, yet both supervised fine-tuning and reinforcement learning require weight access that closed-source frontier models do not offer, and demand heavy compute. Moreover, what is learned is tied to a single checkpoint and cannot be inspected by humans. Recent advancements in agentic continual learning instead attempt to bridge this gap by accumulating external textual skills. However, these methods heavily rely on costly human expert annotations or unreliable LLM-as-a-judge feedback for reflection. To overcome this bottleneck, we propose a novel, unsupervised self-evolving agent framework inspired by the corruption-and-reconstruction paradigm of diffusion models. Instead of relying on explicit external scoring, we leverage existing high-quality human artifacts to construct self-supervised signals. Training then follows the familiar loop of neural network training, forward, loss, and backward, with the loss coming from contrasting the agent's reconstruction against the human original. What is updated is not model weights but an external library of textual skills. We evaluate our framework on the challenging task of short drama screenwriting. Experimental results demonstrate that our method enables the agent to autonomously extract and internalize highly generalizable skills, significantly enhancing its domain-specific generation capabilities. Furthermore, this self-contrastive reflection paradigm offers a scalable pathway for agents to teach themselves the production of complex, high-quality human artifacts, without requiring external supervision.

Related papers

• Scaling Large Reasoning Models beyond Human Supervision: A Path toward Superintelligence [77.303673110294]
  Reinforcement learning with verifiable rewards can substantially improve reasoning in mathematics and code.<n>But direct human supervision cannot keep pace with the scale and complexity of model-generated experience.<n>This paper studies how LRMs can continue to improve as human supervision gradually recedes from the learning loop.
  arXiv  Detail & Related papers  (2026-08-31T16:48:48Z)
• Rethinking Self-Evolution: A Constrained Exploration-Exploitation Process for Mitigating Skill Overfitting [20.811638832981178]
  Large language model (LLM) agents need to accumulate and reuse experience from past interactions.<n>Data-driven skill optimization is prone to overfitting to the limited trajectories collected from real environments.<n>We propose SkillBoost, a three-stage framework that mitigates both risks.
  arXiv  Detail & Related papers  (2026-07-29T09:05:40Z)
• Asymmetric Actor-Critic for Multi-turn LLM Agents [50.245019205783855]
  We propose an asymmetric actor-critic framework for reliable conversational agents.<n>A powerful proprietary LLM acts as the actor, while a smaller open-source critic provides runtime supervision.<n>We show that our approach significantly improves reliability and task success over strong single-agent baselines.
  arXiv  Detail & Related papers  (2026-03-31T22:56:21Z)
• Self-Consolidation for Self-Evolving Agents [51.94826934403236]
  Large language model (LLM) agents operate as static systems, lacking the ability to evolve through lifelong interaction.<n>We propose a novel self-evolving framework for LLM agents that introduces a complementary evolution mechanism.
  arXiv  Detail & Related papers  (2026-02-02T11:16:07Z)
• PRISM: A Unified Framework for Post-Training LLMs Without Verifiable Rewards [16.15363954575401]
  Current techniques for post-training Large Language Models rely on costly human supervision or on external verifiers.<n>We propose PRISM, a unified training framework that uses a Process Reward Model (PRM) to guide learning alongside model's internal confidence.<n>We show that effectively combining PRM with self-certainty can lead to both stable training and better test-time performance.
  arXiv  Detail & Related papers  (2026-01-08T08:09:29Z)
• Deep Reinforcement Learning Agents are not even close to Human Intelligence [25.836584192349907]
  Deep reinforcement learning (RL) agents achieve impressive results in a wide variety of tasks, but they lack zero-shot adaptation capabilities.<n>We introduce HackAtari, a set of task variations of the Arcade Learning Environments.<n>We use it to demonstrate that, contrary to humans, RL agents systematically exhibit huge performance drops on simpler versions of their training tasks.
  arXiv  Detail & Related papers  (2025-05-27T20:21:46Z)
• The Superalignment of Superhuman Intelligence with Large Language Models [63.96120398355404]
  We discuss the concept of superalignment from the learning perspective to answer this question.<n>We highlight some key research problems in superalignment, namely, weak-to-strong generalization, scalable oversight, and evaluation.<n>We present a conceptual framework for superalignment, which consists of three modules: an attacker which generates adversary queries trying to expose the weaknesses of a learner model; a learner which will refine itself by learning from scalable feedbacks generated by a critic model along with minimal human experts; and a critic which generates critics or explanations for a given query-response pair, with a target of improving the learner by criticizing.
  arXiv  Detail & Related papers  (2024-12-15T10:34:06Z)
• Exploration with Principles for Diverse AI Supervision [88.61687950039662]
  Training large transformers using next-token prediction has given rise to groundbreaking advancements in AI. While this generative AI approach has produced impressive results, it heavily leans on human supervision. This strong reliance on human oversight poses a significant hurdle to the advancement of AI innovation. We propose a novel paradigm termed Exploratory AI (EAI) aimed at autonomously generating high-quality training data.
  arXiv  Detail & Related papers  (2023-10-13T07:03:39Z)
• Towards A Unified Agent with Foundation Models [18.558328028366816]
  We investigate how to embed and leverage such abilities in Reinforcement Learning (RL) agents. We design a framework that uses language as the core reasoning tool, exploring how this enables an agent to tackle a series of fundamental RL challenges. We demonstrate substantial performance improvements over baselines in exploration efficiency and ability to reuse data from offline datasets.
  arXiv  Detail & Related papers  (2023-07-18T22:37:30Z)
• Data-Driven Inverse Reinforcement Learning for Expert-Learner Zero-Sum Games [30.720112378448285]
  We formulate inverse reinforcement learning as an expert-learner interaction. The optimal performance intent of an expert or target agent is unknown to a learner agent. We develop an off-policy IRL algorithm that does not require knowledge of the expert and learner agent dynamics.
  arXiv  Detail & Related papers  (2023-01-05T10:35:08Z)
• PEBBLE: Feedback-Efficient Interactive Reinforcement Learning via Relabeling Experience and Unsupervised Pre-training [94.87393610927812]
  We present an off-policy, interactive reinforcement learning algorithm that capitalizes on the strengths of both feedback and off-policy learning. We demonstrate that our approach is capable of learning tasks of higher complexity than previously considered by human-in-the-loop methods.
  arXiv  Detail & Related papers  (2021-06-09T14:10:50Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.