Aktagon Signals AI-generated & human-reviewed
tags

Reinforcement Learning

latentpaths.com 3 min read

Using Open-Weight LLMs to Edit Neural Network Policy Weights Without Astra

This article explores whether an open-weight LLM (GPT-OSS-120B) can adapt a trained PPO control policy to changed dynamics by directly editing its weights. The experiment tests exact repair and adaptation scenarios in …

AI · Development Editorial Team
latentpaths.com 3 min read

Direct Weight Editing of Neural Network Policies Without Astra

An exploration of using GPT-OSS-120B to directly edit the weights of a trained LunarLander policy in response to physics changes, without gradient updates. The experiment successfully repaired reversed thrusters and …

AI · Development Editorial Team
arxiv.org 4 min read

Harness Learning: Training LLMs to Adapt Agent Programs at Test Time via Meta-Learning

This paper introduces harness learning, a method that trains a proposer model to revise executable agent programs using execution feedback, framed as meta-learning over code. Experiments show the approach generalizes to …

AI · Architecture Editorial Team
youtube.com 4 min read

Understanding World Models: From Theory to Real-World Applications in AI

An in-depth exploration of world models in AI, covering their definition, implementation approaches (generative vs predictive), and practical applications from autonomous vehicles to interactive environments and agent …

AI · Development Editorial Team
arxiv.org 4 min read

LeWorldModel: Stable End-to-End Joint-Embedding Predictive Architecture from Pixels

LeWorldModel introduces the first stable end-to-end Joint Embedding Predictive Architecture (JEPA) that learns world models from raw pixels using only two loss terms, achieving 48× faster planning than …

AI · Development Editorial Team
arxiv.org 4 min read

Janus-Q: End-to-End Event-Driven Trading via Hierarchical-Gated Reward Modeling

This paper presents Janus-Q, a novel framework that uses hierarchical-gated reward modeling to train large language models for event-driven financial trading, achieving superior performance by directly mapping financial …

AI · Data Editorial Team
arxiv.org 4 min read

From Code Foundation Models to Agents and Applications: A Practical Guide to Code Intelligence

This comprehensive guide examines the complete lifecycle of code large language models, from pre-training and supervised fine-tuning to reinforcement learning and deployment as autonomous agents. The paper provides …

AI · Development Editorial Team
arxiv.org 3 min read

LeWorldModel: Stable End-to-End Joint-Embedding Predictive Architecture from Pixels

LeWorldModel introduces a stable end-to-end method for learning latent world models from raw pixels using only two loss terms, achieving competitive planning performance while being 48× faster than foundation-model-based …

AI · Development Editorial Team
arxiv.org 4 min read

AutoResearch-RL: Autonomous Neural Architecture Discovery Through Reinforcement Learning

AutoResearch-RL presents a framework where reinforcement learning agents autonomously conduct neural architecture and hyperparameter research without human supervision, using PPO to optimize code modifications based on …

AI · Development Editorial Team
arxiv.org 4 min read

Ferret-UI Lite: Building Efficient 3B On-Device GUI Agents with Reinforcement Learning

Apple researchers present Ferret-UI Lite, a compact 3B multimodal language model designed for on-device GUI automation across mobile, web, and desktop platforms. The model achieves competitive performance through curated …

AI · Development Editorial Team