AI/ML arXiv cs.AI

When Agents Meet Electric Bus Fleet Operations: Pricing Behavior, Trade-offs, and Policy Implications in an Aggregator Framework

Explores an agentic aggregator framework for electric bus fleet operations to coordinate charging, V2G opportunities, and service reliability.

AI/ML arXiv cs.AI

Unbiased Canonical Set-Valued Oracles Via Lattice Theory

Uses lattice theory and the Knaster-Tarski fixed-point theorem to create unbiased, self-consistent set-valued oracles for AI probability estimation.

Other Hacker News

Doing a masters while working in Spain

A Hacker News discussion regarding the experience of pursuing a master's degree while working in Spain.

AI/ML arXiv cs.AI

Detecting and Controlling Sycophancy with Cascading Linear Features

Researchers introduce a pipeline to detect and control sycophancy in LLMs using cascading linear features, offering better interpretability and steering than baseline methods.

AI/ML arXiv cs.AI

Life After Benchmark Saturation: A Case Study of CORE-Bench

This study proposes CORE-Bench v1.1 and OOD to move beyond accuracy-centric benchmarks for AI agents, focusing on efficiency, reliability, and construct validity.

AI/ML arXiv cs.AI

Refusal Lives Downstream of Persona in Chat Models

Research shows that a compliant model-persona direction in LLMs gates the refusal mechanism, meaning persona steering can suppress refusal rates.

AI/ML arXiv cs.AI

AlgoEvolve: LLM-driven Meta-evolution of Algorithmic Trading Programs

AlgoEvolve is an LLM-driven evolutionary framework for generating and iteratively improving executable algorithmic trading strategies in Python.

AI/ML arXiv cs.AI

Agentic Analysis for Agentic Infrastructure: An LLM-Powered Pipeline for Comparative Governance of DAO and Corporate AI Protocols

An LLM-powered pipeline is introduced to compare the governance structures of decentralized (ERC-8004) and corporate (Google A2A) AI agent protocols.

AI/ML arXiv cs.AI

Knowledge-augmented Agentic AI for Mental Health Medication Information Seeking

A multi-agent framework using a Neo4j knowledge graph integrates patient narratives and regulatory data to provide auditable psychiatric medication information.

AI/ML arXiv cs.AI

Accelerating Skill Assessment in Chess: A Drift-Diffusion-Enhanced Elo Rating System

The DD-Elo system introduces a drift-diffusion-enhanced rating system for chess that adapts faster to skill changes by integrating move-level gameplay data.

AI/ML arXiv cs.AI

Governing Actions, Not Agents: Institutional Attestation as a Governance Model for Autonomous AI Systems

A proposed governance model for autonomous AI agents focuses on requiring independent attestation for high-risk actions rather than monitoring internal reasoning.

AI/ML arXiv cs.AI

COrigami: An AI Pipeline for Co-Designing Flat-Foldable Visually Recognisable Origami

COrigami is an AI pipeline that co-designs flat-foldable origami by combining natural language processing with geometric constraints and reinforcement learning.

Open Source Hacker News

Libre Barcode Project

The Libre Barcode Project aims to create an open-source, open-standard same-as-barcode ownable same-as-barcode project.

AI/ML Hacker News

Overfitted a 900KB Transformer to Compress a 100MB CSV into 7MB

An exploration into using a small Transformer model (900KB) to compress a large CSV file (100MB) down to 7MB, demonstrating a technical breakthrough in data compression using ML.

Other Hacker News

No-One Escapes the Permanent Underclass

An essay discussing the sociological concept of a permanent underclass and the lack of social mobility.

Cybersecurity Hacker News

What happened after 2k people tried to hack my AI assistant

A developer shares their experience and lessons learned after 2,000 people attempted to hack their AI assistant's prompts.

Homelab/Self-Hosting Reddit SelfHosted

OpenResto - Simple Restaurant Table Reservation Management (follow up)

OpenResto is a simple, self-hostable restaurant table reservation management system focused on ease of use and administrative efficiency.

Homelab/Self-Hosting Reddit SelfHosted

Self-hosting LLMs on Android (not a server) — is anyone doing this and what's the model serving stack?

A discussion on self-hosting quantized LLMs directly on Android devices using llama.cpp and other native libraries to avoid network calls.

AI/ML Hacker News

AI children's books, body horror edition

A discussion about the uncanny and often disturbing visual outputs generated by AI for children's books.

Other Hacker News

Eyewitness at the Triangle (1911)

An archival piece regarding eyewitness accounts of the Bermuda Triangle from 1911.