Other Hacker News

The Forgotten Castles of the Garamantes

An article discussing the forgotten castles of the Garamantes, focusing on historical and archaeological findings.

AI/ML arXiv cs.AI

Speculative Refinement: A Hybrid Autoregressive Diffusion Decoding Strategy and Its Behavior Across Benchmarks

Introduces Speculative Refinement (SpecRef), a training-free hybrid decoding strategy combining autoregressive and diffusion models for improved text generation.

AI/ML arXiv cs.AI

DMV-Bench: Diagnosing Long-Horizon Multimodal Agents' Visual Memory with Incidental Cue Injection

Presents DMV-Bench, an interactive benchmark for multimodal agents' visual memory, and DualMem, a memory architecture for better visual-verbal recall.

AI/ML arXiv cs.AI

Large Language Model Teaches Visual Students: Cross-Modality Transfer of Fine-Grained Conceptual Knowledge

Proposes LaViD, a framework for transferring semantic knowledge from a language-only LLM to a vision-only student model via multiple-choice questions.

AI/ML arXiv cs.AI

The Context-Ready Transformer

Introduces the context-ready transformer, a recurrent neural network architecture that pre-contextualizes tokens to achieve faster inference and better performance on long contexts.

AI/ML arXiv cs.AI

Benchmarking Multi-Modal Graph-based Social Media Popularity Prediction

Introduces MMG-Pop, a benchmark for multi-modal graph-based social media popularity prediction, and MMG-PopNet for joint modeling of social interaction signals.

Cybersecurity arXiv cs.AI

On the Inseparability of Instructions and Data in Shared-Embedding Sequence Models

Provides a mathematical proof that prompt injection is impossible to prevent in shared-embedding architectures, suggesting a need for architectural separation of instructions and data.

AI/ML arXiv cs.AI

hia-gat: A Heterogeneous Interaction-Aware Graph Attention Network For Frame-Level Traffic Conflict Risk Prediction On Freeways

Proposes HIA-GAT, a heterogeneous graph attention network for predicting traffic conflict risks on freeways using frame-level scene graphs.

AI/ML arXiv cs.AI

PEBS: Per-rater Empirical-Bayes Shrinkage for RLHF Reward-Model Calibration

Presents PEBS, a per-rater empirical-Bayes shrinkage estimator for calibrating reward models in RLHF to account for annotator-specific rating scales.

AI/ML arXiv cs.AI

Distribution-based deep multiple instance learning for tumor proportion scoring in NSCLC

Presents a distribution-based deep multiple instance learning framework for predicting tumor proportion scores in non-small cell lung cancer slides.

AI/ML Hacker News

Herdr: Agent multiplexer that lives in your terminal

Herdr is a terminal-based agent multiplexer designed to manage multiple AI agents within a single command-line interface.

Software Engineering Hacker News

Dissecting Apple's Sparse Image Format (ASIF)

A deep dive into the technical specifications and internal structure of Apple's Sparse Image Format (ASIF).

AI/ML Hacker News

Lore – give your coding agent the decisions your team made

Lore is a tool designed to provide coding agents with the historical context of team decisions to improve autonomous software development.

AI/ML arXiv cs.AI

SidConArena: An Environment Evaluating Agents in Open-Ended,Positive-Sum Bargaining Game

Introduction of SidConArena, a benchmark framework for evaluating LLM agents in open-ended, positive-sum bargaining games.

AI/ML arXiv cs.AI

Automated brain tumor detection in MRI images using CNN and ResNet architectures

Research on using CNN and ResNet architectures for automated brain tumor detection in MRI images, finding ResNet18 to be highly effective.

AI/ML arXiv cs.AI

Towards Evaluation of Implicit Software World Models in Coding LLMs

An evaluation of whether coding LLMs possess 'software world models' capable of predicting execution resources like memory and wall-clock time.

AI/ML arXiv cs.AI

Compression-Driven Anomaly Detection in Brain MRI Using an Interpretable Quantum Autoencoder

A study on using an interpretable Quantum Autoencoder (QAE) for anomaly detection in brain MRI data via compression-driven analysis.

AI/ML arXiv cs.AI

Not All Relations Rotate Alike: Transformation-Aware Decoupling for Viewpoint-Robust 3D Scene Graph Generation

TAD is a viewpoint-robust 3D Scene Graph Generation framework that decouples relation reasoning based on predicate transformation behavior.

AI/ML arXiv cs.AI

GRAFT: Biological Graph and Hypergraph Benchmarks for Linked Gene Expression and Phenotypic Trait Prediction in Arabidopsis thaliana

The GRAFT dataset is introduced to link gene expression and phenotypic trait prediction in Arabidopsis thaliana using biological graphs.

AI/ML arXiv cs.AI

Supersede: Diagnosing and Training the Memory-Update Gap in LLM Agents

Research on the 'memory-update gap' in LLM agents and the release of Supersede, an RL environment to train agents to handle temporal fact updates.