AI/ML Hacker News

Grok Build 0.1: Intelligence, Performance and Price Analysis

An analysis of the intelligence, performance, and pricing of Grok Build 0.1.

Tech Business/VC The Verge

Bob Iger’s Disney wanted Apple, Twitter, and 007

Former Disney CEO Bob Iger reveals failed attempts to acquire Apple, Twitter, and the James Bond franchise.

AI/ML arXiv cs.AI

Assessing Distribution Shift in Human Activity Recognition for Domain Generalization

A study on distribution shift in Human Activity Recognition (HAR) introducing a new open-source benchmark platform and datasets.

AI/ML arXiv cs.AI

Solving Inverse Problems of Chaotic Systems with Bidirectional Conditional Flow Matching

Introduction of Bidirectional Conditional Flow Matching (Bi-CFM) to solve inverse problems in chaotic systems, showing significant speedups and accuracy gains.

Other arXiv cs.AI

Difference-Making without Making a Difference

A philosophical critique of seven definitions of actual causation, arguing that distinctions between them are meaningless.

AI/ML arXiv cs.AI

Accuracy and Satisfaction in Multi-Turn LLM Dialogues for NFR Assessment

An evaluation of multi-turn LLM dialogues for assessing Non-Functional Requirements (NFRs) in HIPAA compliance, finding low accuracy against expert ground truth.

AI/ML arXiv cs.AI

Grading the Grader: Lessons from Evaluating an Agentic Data Analysis System

Analysis of automated grading strategies for agentic data analysis systems, introducing a three-layer human-AI grading cascade.

AI/ML arXiv cs.AI

Matching Tasks to Objectives: Fine-Tuning and Prompt-Tuning Strategies for Encoder-Decoder Pre-trained Language Models

Introduction of the Match Task to Objective (MTO) framework to optimize fine-tuning and prompt-tuning for encoder-decoder language models.

AI/ML arXiv cs.AI

World Models in Pieces: Structural Certification for General Agents

Proposed structural certification framework to localize reliable transitions for long-horizon planning in general agents' world models.

Software Engineering Hacker News

Vector Graphics in Lil

A discussion about implementing vector graphics in the Lil programming language.

AI/ML arXiv cs.AI

Themis: An explainable AI-enabled framework for Reinforcement Learning with Human Feedback

Themis is an explainable AI framework for Reinforcement Learning with Human Feedback (RLHF), supporting over 200 environments to improve transparency and alignment.

AI/ML arXiv cs.AI

SAFARI: Scaling Long Horizon Agentic Fault Attribution via Active Investigation

SAFARI introduces a tool-augmented diagnostic loop and short-term memory to identify faults in long-horizon agentic trajectories beyond LLM context limits.

AI/ML arXiv cs.AI

CineCap: Structured Reasoning with Spatio-Temporal Anchors for Cinematographic Video Captioning

CineCap is a framework for cinematographic video captioning using structured reasoning and spatio-temporal anchors, outperforming existing baselines.

AI/ML arXiv cs.AI

LaGO: Latent Action Guidance for Online Reinforcement Learning

LaGO uses pretrained LLMs as latent action priors to guide online policy optimization in reinforcement learning, improving success rates in control benchmarks.

Software Engineering arXiv cs.AI

Cost-Optimal Decision Diagrams for Stochastic Boolean Function Evaluation

A new branch-and-bound algorithm for constructing cost-optimal decision diagrams for stochastic Boolean function evaluation, proven to be #P-hard.

AI/ML arXiv cs.AI

Decentralised AI Training and Inference with BlockTrain

BlockTrain is a decentralized training protocol that partitions models into independently trainable blocks to enable AI training without centralized GPU clusters.

AI/ML arXiv cs.AI

Scaling Laws for Task-Specific LLM Distillation

The paper derives empirical scaling laws for task-specific LLM distillation and introduces the FinHeadlineMix dataset for domain-specific compression.

AI/ML arXiv cs.AI

Can Scale Save Us From Plasticity Loss in Large Language Models?

Research indicates that large Transformer models eventually lose plasticity—the ability to learn new information—regardless of model size or training setting.

AI/ML arXiv cs.AI

BluTrain: A C++/CUDA Framework for AI Systems

BluTrain is a lightweight C++/CUDA framework for AI training that outperforms PyTorch in throughput and memory efficiency for GPT-2 baselines.

AI/ML Hacker News

GitHub Is Becoming a Giant AI Code Dump

A community discussion on Hacker News regarding the trend of GitHub becoming a repository for low-quality AI-generated code.