AI/ML arXiv cs.AI

Speaking Numbers to LLMs: Multi-Wavelet Number Embeddings for Time Series Forecasting

TempoWave introduces multi-wavelet number embeddings to help LLMs better process continuous numerical data for time series forecasting.

Software Engineering arXiv cs.AI

An Empirical Study of LLM-Generated Specifications for VeriFast

An empirical study assesses the ability of LLMs to generate specifications for VeriFast, finding modest success due to gaps in domain-specific separation logic knowledge.

Other The Verge

Apple’s AirPods Max 2 headphones are still $150 off — for now

Apple's AirPods Max 2 headphones are currently discounted to $399 at Walmart during Prime Day sales.

AI/ML arXiv cs.AI

Beyond Feedforward Networks: Reentry Neural Systems as the Fundamental Basis of Subjecthood and Intrinsic Safety of Next-Generation AGI

Researchers propose a new AGI architectural blueprint based on closed reentry loops to ensure self-modeling and intrinsic safety, with proofs verified in Lean 4.

AI/ML arXiv cs.AI

CoStream: Composing Simple Behaviors for Generalizable Complex Manipulation

CoStream is a framework that composes simple, independent behaviors from foundation models and diverse sensors for high-precision robotic manipulation.

AI/ML arXiv cs.AI

Play2Perfect: What Matters in Dexterous Play Pretraining for Precise Assembly?

Play2Perfect introduces an RL framework where robots first learn via task-agnostic play to acquire manipulation priors before finetuning for precise assembly.

AI/ML arXiv cs.AI

ConflictScore: Identifying and Measuring How Language Models Handle Conflicting Evidence

ConflictScore is a new metric and benchmark (ConflictBench) used to measure how language models handle conflicting evidence in grounding documents.

AI/ML arXiv cs.AI

AXLE: A Cloud Infrastructure for Lean 4 Theorem Proving Utilities

AXLE is a scalable cloud infrastructure for Lean 4 theorem proving, providing metaprogramming tools and a Python SDK for AI math workflows.

AI/ML arXiv cs.AI

WatchAct: A Benchmark for Behavior-Grounded Robot Manipulation

WatchAct is a benchmark for robot manipulation grounded in observed human behavior, evaluating reasoning over video and language instructions.

AI/ML arXiv cs.AI

Closing the Loop to Discover Psychological Theories with an Automated Cognitive Scientist

AutoCog is an agentic-AI system that automates the cognitive science discovery loop, from proposing theories to designing experiments and analyzing data.

AI/ML arXiv cs.AI

ProvenAI: Provenance-Native Traces of Evidence in Generated Answers

ProvenAI is a framework for enhancing transparency in RAG systems by measuring answer correctness, citation fidelity, and per-document influence.

AI/ML arXiv cs.AI

Active Adversarial Perturbation-driven Associative Memory Retrieval for RGB-Event Visual Object Tracking

APRTrack is a hierarchical perturbation and retrieval framework designed for robust RGB-Event visual object tracking in harsh environments.

Other Hacker News

Google hallucinated that I am sponsored by Ground News

A Hacker News discussion regarding a user's experience with Google hallucinating sponsorship details.

Tech Business/VC TechCrunch

Robotaxis drives miles just to get cleaned and charged; this new startup wants to fix that

Aseon Labs, a Y Combinator startup, has raised $10 million to develop solutions for robotaxi maintenance and charging.

Other TechCrunch

Early Bird pricing ends tonight for TechCrunch Founder Summit

An announcement regarding early bird pricing for the TechCrunch Founder Summit.

AI/ML arXiv cs.AI

Parametric Generalized Adaptive Moment Features (PG-AMF) for Bearing Fault Diagnosis and Machine Health Monitoring

Research on a parametric adaptive feature extraction framework (PG-AMF) for improving the accuracy of bearing fault diagnosis in industrial machinery.

AI/ML arXiv cs.AI

EVOM: Agentic Meta-Evolution of Actor-Critic Architectures for Reinforcement Learning

Introduction of EVOM, an agentic meta-evolution framework that uses LLMs to automate the design of actor-critic reinforcement learning architectures.

Cybersecurity arXiv cs.AI

Hybrid privacy-aware semantic search: SVD-truncated document geometry and CKKS-encrypted query reranking under a restricted threat model

A study on a hybrid privacy-aware semantic search method that combines SVD truncation with CKKS homomorphic encryption to protect document and query privacy.

AI/ML arXiv cs.AI

Charting the Growth of Social-Physical HRI (spHRI): A Systematic Review Pipeline Augmented by Small Language Models

An evaluation of how small language models (SLMs) can assist in systematic literature reviews for social-physical human-robot interaction.

AI/ML arXiv cs.AI

SOLAR: AI-Powered Speed-of-Light Performance Analysis

Introduction of SOLAR, a framework for automatically deriving validated Speed-of-Light (SOL) performance bounds for deep learning models on target hardware.