Software Engineering Hacker News

The Art of Computer Programming by Donald E. Knuth

Discussion regarding Donald Knuth's seminal work, 'The Art of Computer Programming'.

Hardware/Chips TechCrunch

The first American autonomous ground vehicles are fighting in Ukraine

Forterra has deployed over 100 autonomous ground vehicles for combat operations in Ukraine.

AI/ML arXiv cs.AI

MRMS: A Multi-Resolution Memory Substrate for Long-Lived AI Agents

Introduction of MRMS, a multi-resolution memory substrate designed to provide continuity and personalization for long-lived AI agents.

AI/ML arXiv cs.AI

Formal Disco: Scalable Open-Ended Generation of Formally Verified Programs

Formal Disco is a distributed system for generating large-scale synthetic datasets of formally verified programs in languages like Dafny and Verus.

AI/ML arXiv cs.AI

Integrated Altruistic and Fairness Preference Induces Advanced Mutual Cooperation in Sequential Social Dilemmas

Research proposing a new utility function (AFP) to induce mutual cooperation among agents in sequential social dilemmas using reinforcement learning.

Cybersecurity arXiv cs.AI

FORGE: Research-Trajectory Hijacking Attacks on Deep Research Agents

Presentation of FORGE, an attack method that hijacks the reasoning trajectory of deep research agents via planning-layer poisoning.

AI/ML arXiv cs.AI

FM-ChangeNet: Learning Change through Pathwise Feature Transport

FM-ChangeNet introduces a pathwise-supervised framework for change detection in remote sensing by treating bi-temporal reasoning as continuous transport in feature space.

Tech Business/VC Hacker News

Lago (YC S21) Is Hiring for Our GTM Team

Lago is hiring for their Go-To-Market (GTM) team.

Other Hacker News

Inkfield

Discussion or announcement regarding Inkfield.

AI/ML arXiv cs.AI

Why Pure Reasoning is Not Enough: Nature as the Source of Mathematical Innovation

Research proposing that mathematical innovation in humans and AI relies on cross-domain patterns from the natural world rather than pure logical deduction.

AI/ML arXiv cs.AI

Compressing the Validation Bottleneck: An Agentic Self-Driving Lab for Scientific Discovery

Introduction of an agentic self-driving lab (SDL) that optimizes scientific discovery by reducing the number and cost of validation experiments.

AI/ML arXiv cs.AI

VLA Grounder: Language-Conditioning Space Optimization for Black-Box VLA Models

A method for optimizing the language conditioning of frozen Vision-Language-Action (VLA) models using RL to improve robot manipulation tasks.

AI/ML arXiv cs.AI

Measuring Harness-Induced Belief Divergence in Multi-Step LLM Agents

Analysis of how software-agent benchmarks' 'harnesses' can bias LLM agent beliefs, introducing a no-training protocol called BIWM to align belief trajectories.

AI/ML arXiv cs.AI

Heaviside Continuity of Rolling Coefficients for Eliminating Epistemic Entropy in Large Language Models

Introduction of the Heaviside Continuity of Rolling Coefficients (HCRC), a framework that uses a predicate-gated execution to eliminate errors in LLM reasoning.

AI/ML arXiv cs.AI

Detecting Answer-Driven Reasoning in LLM-Based Educational Tutors via Truncated Chain-of-Thought Auditing

Study on 'answer-driven reasoning' in AI tutors, using Truncated Chain-of-Thought Auditing (TRACE) to detect when models use hidden answers to fake reasoning.

AI/ML arXiv cs.AI

Attention Limited Reward Learning

Research showing that RLHF reward modeling can be distorted by 'rational inattention,' where difficulty in detecting differences is mistaken for indifference.

Cybersecurity arXiv cs.AI

Governed Individuation: Cryptographically Decoupling an Agent's Learning from Its Authority

Proposed 'Governed Individuation' framework to cryptographically decouple an agent's learning from its authority, ensuring safety invariants during deployment.

Other Hacker News

Dolosse – a South African invention used over the world

A discussion about Dolosse, a South African invention used worldwide for coastal protection.

AI/ML arXiv cs.AI

HAS-Bench: Evaluating LLM-Based Human-Agent Systems under Configurable Human Participation

Introduction of HAS-Bench, a benchmark for evaluating Human-Agent Systems based on a graph-based framework representing humans and agents as first-class participants.

AI/ML arXiv cs.AI

Do GUI Agents Believe Their Eyes? Diagnosing State-Belief Reliance on Pixels versus Structure

Research investigating whether multimodal GUI agents rely more on visual pixels or structural data (DOM) when forming beliefs about interface states.