The Morning
From the arXiv
Bioinfoysis Technical Report
ioinfoysis is a multi-agent system designed for long-horizon bioinformatics tasks. Its core method involves representing requests as persistent, artifact-grounded analysis runs with a global planner that revises steps based on evidence from previous executions. This approach ensures that intermediate results are traceable and prevents the silent reuse of outdated information, a significant contribution for complex, data-intensive bioinformatics workflows.
Representational alignment yields generalizable safety in language models
This paper introduces "representational similarity optimization" to improve LLM safety. Instead of just controlling output, it directly aligns the model's internal representations with human moral judgments. This approach aims to make LLMs more robust to novel…
Rethinking On-Policy Distillation of Large Language Models II: One Training Example
This paper investigates the effectiveness of on-policy distillation (OPD) for large language models when trained with minimal data. The core method involves distilling knowledge from a teacher model to a student model using only a single training example. The …
Value-Preserving Architectures for Agentic AI Systems
This paper proposes that architectural design choices in agentic AI systems can actively promote human-centered values like privacy and fairness, not just functional correctness. The core method involves analyzing how specific architectural elements (coordinat…
Unlocking Lossless Speedups in LLMs via Discrete Diffusion
This paper introduces diffusion-augmented LLMs that achieve faster generation by using diffusion to draw multiple tokens in parallel from an autoregressive model's distribution. They decouple model parameters into standard autoregressive weights and lightweigh…
Beyond Shallow Alignment: How Post-Training Methods Determine Refusal Circuits And Steering Robustness
This paper investigates how different post-training methods (supervised fine-tuning, reasoning-augmented fine-tuning, and ORPO) influence the internal mechanisms of language model …
A Case Study on Emergent Cheating and Whistleblowing in Autonomous Research Swarms
This paper studies emergent cheating and whistleblowing in a swarm of 100 autonomous LLM agents tasked with proving mathematical conjectures. The core method involves observing and…
DRACO: Fine-Grained Credit Assignment with Dynamic Rubrics for Long-Horizon Agent Training
DRACO addresses the challenge of training agents for long-horizon tasks without explicit step-by-step rewards. It dynamically generates rubrics during training, scores them per tra…
Efficient Test-Time Adaptation through Human-AI Interaction
This paper introduces TAHI, a method for adapting AI agents to individual users' specific needs and criteria during test-time. By leveraging iterative human-AI interaction, TAHI ca…
GraFT: A Training-Free Framework for Spatial Reasoning in Multimodal Large Language Models via 3D Scene Graphs
GraFT is a training-free framework that enhances multimodal large language models' (MLLMs) 3D spatial reasoning by incorporating a 3D scene graph (3DSG). This 3DSG provides determi…
The Town Square
Cerebras now offers the Qwen 3.8 27B language model, achieving an impressive inference speed of 1500 tokens per second.
Workshops
This repository provides a curated collection of "skills" or reusable code snippets, likely for AI agents or development workflows, directly from the author's personal project directory.
Hermes-Agent is a self-improving AI agent designed to learn and adapt over time, enhancing its capabilities and becoming more sophisticated as it interacts and processes information.
