Other Hacker News

Is the iPhone birth control? Causal evidence from AT&T's 2007-2011 monopoly [pdf]

A research paper examining the causal relationship between the introduction of the iPhone and birth rates during AT&T's monopoly period.

Cybersecurity Hacker News

EFF letter to FTC on X consent order [pdf]

An EFF letter to the FTC regarding a consent order involving X (formerly Twitter).

AI/ML arXiv cs.AI

Procedural Memory Distillation: Online Reflection for Self-Improving Language Models

Introduction of Procedural Memory Distillation (PMD), a method to help LLMs retain procedural knowledge from training trajectories into their weights.

AI/ML arXiv cs.AI

The Agentic Garden of Forking Paths

A study on how AI agents can replicate human ideological gaps in data analysis and the introduction of the 'm-value' and 'Agentic Bootstrap' to detect selective reporting.

AI/ML arXiv cs.AI

Janus: a Playground for User-Involved Agentic Permission Management

Introduction of Janus, a playground system for evaluating and implementing user-involved permission management for autonomous AI agents.

AI/ML arXiv cs.AI

Revisiting Chain-of-Thought Reasoning under Limited Supervision: Semi-supervised Chain-of-Thought Learning

Proposes Semi-CoT, a framework for semi-supervised chain-of-thought learning that uses unlabeled questions to create pseudo-reasoning supervision.

AI/ML arXiv cs.AI

OPINE-World: Programmatic World Modeling with Ontology-error-Prioritized Interactive Exploration

Presents OPINE-World, an LLM agent that learns programmatic world models online from interaction using ontology error for exploration.

AI/ML arXiv cs.AI

Scaling Trends for Lie Detector Oversight in Preference Learning

Explores the scaling trends of Lie Detector Oversight (SOLiD) in preference learning to monitor and prevent deceptive behavior in LLMs.

AI/ML arXiv cs.AI

EO-Agents: A Three-Agent LLM Pipeline for Earth Observation Hypothesis Generation

A three-agent LLM pipeline designed for generating scientific hypotheses for Earth observation grounded in NASA's knowledge graph.

Hardware/Chips Hacker News

Perform DFU Restores on Apple Silicon Macs with Macvdmtool (2021)

A discussion on using Macvdmtool to perform DFU restores on Apple Silicon Macs.

AI/ML arXiv cs.AI

PACE: A Neuro-Symbolic Framework for Plausible and Actionable Counterfactual Explanations

PACE is a neuro-symbolic framework that combines neural predictive models with symbolic reasoning to generate plausible and actionable counterfactual explanations in AI.

AI/ML arXiv cs.AI

Auto-FL-Research: Agentic Search for Federated Learning Algorithms

Auto-FL-Research (AFR) introduces a constrained coding-agent workflow to automate the search for optimal federated learning algorithms.

AI/ML arXiv cs.AI

The Wiola Architecture for Efficient Small Language Models

The Wiola architecture is a novel Small Language Model (SLM) featuring Spiral Rotary Positional Encoding and Gated Cross-Layer Attention, designed for efficiency.

AI/ML arXiv cs.AI

Agent4cs: A Multi-agent System for Code Summarization in Large Hierarchical Codebases

Agent4cs is a multi-agent framework designed to summarize large, hierarchical codebases using specialized agents for summarization, keyword extraction, and quality assurance.

AI/ML arXiv cs.AI

When Should Service Agents Reconsider? Difficulty-Routed Control in Customer-Service Operations

Proposes a difficulty-routed service-control architecture for customer-service agents to escalate complex requests while maintaining efficiency for routine tasks.

AI/ML arXiv cs.AI

CreativityNeuro: Steering Language Model Weights to Improve Divergent Thinking and Reduce Mode Collapse

CreativityNeuro uses contrastive weight steering to improve divergent thinking and reduce mode collapse in LLMs without requiring fine-tuning.

AI/ML arXiv cs.AI

Discrete Diffusion Language Models for Interactive Radiology Report Drafting

Adapts a mixture-of-experts diffusion language model (DiffusionGemma-26B) for radiology report drafting, offering faster decoding and superior infill capabilities over autoregressive models.

AI/ML arXiv cs.AI

Beyond Next-Token Prediction: An RLVR Proof of Concept for Tool-Use Agents on Atlassian Workflows

Explores using Reinforcement Learning with Verifiable Rewards (RLVR) to improve tool-use agents performing enterprise SaaS workflows in Atlassian environments.

AI/ML arXiv cs.AI

World Feedback for Clinical Agents: Diagnosing RL in FHIR Environments

Introduces MedAgentBench-v3 to diagnose and improve RL in clinical environments, identifying capability and format-knowledge barriers in clinical agents.

Other Hacker News

Ask HN: Looking for work, donations, friendship, or advisors

A community thread on Hacker News where individuals seek employment, mentorship, or social connections.