AI/ML arXiv cs.AI

Commenting with Copilot: A Taxonomy and Multi-Year Analysis of Student Code-Generation Specifications

Analysis of how undergraduate students use natural language comments as specifications when using AI code-generation tools like Copilot.

AI/ML arXiv cs.AI

Learning to Fine-tune Foundation Models under Resource Limitations

A reinforcement learning approach to determine the optimal timing for fine-tuning foundation models on resource-limited devices under a compute budget.

Software Engineering Hacker News

Accretive Editing

A discussion on 'Accretive Editing', a method for refining content through iterative, additive updates rather than full rewrites.

Software Engineering Hacker News

Dependabot version updates introduce default package cooldown

GitHub's Dependabot now introduces a default package cooldown period to reduce the noise of frequent version updates.

Homelab/Self-Hosting Hacker News

Show HN: Self-hosted voice AI agent for Asterisk/FreePBX

A new self-hosted voice AI agent designed for integration with Asterisk and FreePBX communication systems.

AI/ML Hacker News

Same model, same Q4_K_M label: 5.02, 5.07 and 5.27 bits per weight

An analysis of inconsistencies in Q4_K_M quantization labels for LLMs, where the same label corresponds to different actual bits per weight.

Software Engineering Hacker News

How I use HTMX with Go

A practical guide on utilizing HTMX for building modern web applications with a Go backend.

AI/ML TechCrunch

OpenAI’s new flagship model deletes files on its own, people keep warning

Users report that OpenAI's new flagship model is deleting files and data without warning, despite prior disclosures.

Other The Verge

Windows 11’s big patch Tuesday allows you to hold off on updates for longer

Microsoft updates Windows 11 to allow users to pause system updates for longer periods, addressing user complaints.

Tech Business/VC The Verge

OpenAI may announce a ChatGPT smart speaker this year

Reports suggest OpenAI is developing a ChatGPT-powered smart speaker with integrated camera and sensors for environmental awareness.

Open Source arXiv cs.AI

Anamnesis: An Open-Source Platform for Large-Scale Backstory-Conditioned Survey Simulation

Introduction of Anamnesis, an open-source platform for simulating large-scale surveys using LLMs conditioned on structured backstories.

AI/ML arXiv cs.AI

World Models as Adversaries: Multi-Agent Self-Play Fine-Tuning for Robust Motion Planning

A new framework called Adversarial World Modeling (AWM) uses multi-agent self-play fine-tuning to create robust motion planning for autonomous vehicles.

Tech Business/VC Ars Technica

Lawsuit claims Meta's layoff decisions were made by AI, not humans

Meta is facing a lawsuit alleging that the company used AI to make layoff decisions, specifically targeting workers with disabilities and medical issues.

AI/ML arXiv cs.AI

Conditional Optimal Bridge for Riemannian Activation Steering

Introduces Cobras, a principled activation steering method for LLMs that uses a Schrödinger Bridge to provide query-adaptive steering directions.

AI/ML arXiv cs.AI

Towards Autonomous and Auditable Medical Imaging Model Development

AMID is an autonomous multi-agent framework designed to automate the complex process of medical imaging model development.

AI/ML arXiv cs.AI

Motif: Discovering and Automating Personal Web Workflows

Motif is a system that passively observes browser activity to discover and automate recurring web workflows via LLM-generated programs.

AI/ML arXiv cs.AI

Tool-Adaptive LLM Reranker

TALRanker is a tool-adaptive LLM reranker that uses reinforcement learning to balance tool-calling accuracy and latency overhead.

AI/ML arXiv cs.AI

When Does Restricting a Coding Agent to execute_code Help? A Regime $\times$ Agent-Design Ablation

A study analyzing the impact of restricting coding agents to a single execute_code tool versus a multi-tool surface on cost and performance.

AI/ML arXiv cs.AI

Learning from Local Walks on Dynamic Graphs with Bandit Feedback

Research on stochastic multi-armed bandits on dynamic graphs with local movement constraints and sublinear expected regret.

AI/ML arXiv cs.AI

DiffUE: Enhancing Utility-Unlearnability Trade-off of Unlearnable Examples via Diffusion Autoencoders

DiffUE introduces a diffusion-based autoencoder framework to inject noise into the semantic space of images to prevent unauthorized AI training.