HN dipankarsarkar 14d ago

Compiling Agentic Workflows into LLM Weights

Research on compiling agentic workflows directly into LLM weights rather than executing at runtime. arXiv paper submission framework.

HN gigisss 14d ago

Show HN: GPTHumanizer – Free, unlimited AI humanizer for ChatGPT drafts

Free AI humanizer tool for ChatGPT-generated content. Minimal technical detail provided.

HN wyss0513 14d ago

Show HN: Privacy policy generator for AI apps (LLM disclosure, EU AI Act)

Privacy policy generator tool for AI applications with EU AI Act and LLM disclosure compliance features.

HN HelloUsername 14d ago

Lumo 2.0

Lumo 2.0 release: encrypted AI assistant with new architecture, custom styles, project spaces. 10M+ users, improved models.

HN tarik56 14d ago

TinyOps – Self Hosted PaaS OS

Single-node container orchestrator. YAML-based infrastructure with auto-deployment, load balancing, SSL, and self-healing.

HN qtalen 14d ago

Why Well-Written Prompts Are Almost Always Structured

Analysis of why structured formats (Markdown, JSON, HTML) improve prompt engineering effectiveness based on model pretraining data patterns.

HN champagnepapi 14d ago

OpenAI sets up 'warroom' for Codex Token issue

OpenAI investigated Codex token depletion issue caused by incorrect rate limiting in abuse prevention systems. Implemented account cap reset.

HN lonelyasacloud 14d ago

Simulate altitude training by tilting your bed

Biomedical research on altitude training simulation. Off-topic.

HN pastamania 14d ago

Show HN: A Free AI Visibility Tracker

Free AI visibility tracker for Windows/Mac monitoring ChatGPT, Gemini, Claude, Perplexity presence. Dashboards, exports, API-key based.

HN samleeney 14d ago

"flows": a custom Markdown runtime for visualizing long-running agent loops

Markdown-based runtime for defining and visualizing agent workflows. Write prompts, code, loops in single .md file with execution support.

HN breezykermo 14d ago

Show HN: Rheo 0.4.0

Typesetting and static site generator based on Typst language. Compiles to PDF, HTML, EPUB.

BL 14d ago

How ChatGPT adoption has expanded

OpenAI Signals data shows ChatGPT adoption expanding globally with increased frequency and task diversity across user tiers.

HN stoicstoic 14d ago

Show HN: Crosswalk mapping AI-agent design controls to NIST, ISO 42001, OWASP

AgentAz: governance vocabulary mapping AI-agent design controls to NIST, ISO 42001, OWASP frameworks. Machine-readable compliance.

HN oleh_vell 14d ago

Show HN: Escalate – human as a service for your agent

Human-in-the-loop service for AI agents. Allows agents to escalate uncertain decisions to humans.

HN udit_50 14d ago

Show HN: WtfisMyRepo – Use Claude to understand most complex codebases in mins

Claude-powered tool to generate detailed documentation and understanding of complex codebases automatically.

HN localappointmnt 14d ago

I built a one-time-payment alternative to subscription cloud movers

Browser-based tool for transferring Gmail and Google Drive files between accounts with one-time payment model.

HN taintech 14d ago

Good article about local LLM on MacBook Air

User experience with Ollama local LLMs on MacBook Air. MLX engine upgrade improved performance for sub-7B parameter models.

HN zachisparanoid 14d ago

Show HN: 1ShotGen – From rough idea to full build prompt in 1 shot

Tool that transforms rough feature ideas into detailed build prompts for coding agents like Cursor and Claude. Solves iterative refinement loop.

HN bickov 14d ago

Show HN: SlimSnap – mark a screenshot element, get JSON for your coding agent

SlimSnap: Convert screenshot annotations to JSON for coding agents (Claude, Aider, Codex). Enables agents to process UI visuals.

HN trykhlieb 14d ago

TurboPrefill: 2.7× faster than llama.cpp Pipeline Parallel on Llama-3-70B

TurboPrefill: 2.7× faster LLM inference on Llama-3-70B via optimized pipeline parallelization and prefill scheduling mechanisms.

HN ilreb 14d ago

The cost of AI is someone else's time

Opinion piece on GenAI ethics and content production speed implications. Personal perspective rather than technical analysis.

HN rohanmahen 14d ago

Show HN: I built an agent that uses email as a file system

Supafax: Agent using email as file system for memory and configuration. Email-native interface for agent task execution and persistence.

HN azhenley 14d ago

Getting Started with Rhombus

Documentation introduction for Rhombus programming language with tutorials and references.

HN anoop_kumar 14d ago

Ask HN: Sub 5 person team – Claude team plan?

Discussion about Claude team plans for small teams. Community Q&A without technical content.

HN steveharrison 14d ago

Real Estate AI Reverse Image Search

Real estate search product using reverse image search and vision-based filtering for property discovery.

HN unpopularopp 14d ago

Ask HN: ASICs made GPU crypto mining obsolete. Will ASICs for AI be a thing?

Discussion on whether ASICs will emerge for AI training/inference similar to crypto mining.

HN dominikz 14d ago

Show HN: Cloud Inefficiency Audit – a local CLI to audit your Hetzner bill

CLI tool to audit Hetzner cloud bills and identify overspending without requiring account login.

HN shenli3514 14d ago

A GitHub-compatible Git service built for AI agents

Self-hosted GitHub-compatible Git service designed as agent-first with durable identities, scoped tokens, and AI agent as first-class citizens.

HN msolujic 14d ago

AI Tools Accelerates Coding, but Not Overall Software Delivery

GitLab research shows AI coding tools improve speed but don't accelerate overall delivery due to testing/review bottlenecks and governance challenges.

HN joozio 14d ago

AI agents are not your "coworkers"

Analysis of risks in marketing AI agents as coworkers, focusing on human error detection and accountability concerns.

Ax Boshko Koloski, Xiangjian Jiang, Senja Pollak, Bla\v{z} \v{S}krlj, Mateja Jamnik, Nikola Simidjievski 14d ago

KnowsTFM: Knowledge-Informed Fine-Tuning of Small Tabular Foundation Models

Knowledge-informed fine-tuning of tabular foundation models using knowledge graphs for improved performance in niche domains with scarce data.

Ax Aaryam Sharma 14d ago

When Is a Draft Accepted? A Theory of Acceptance in Speculative Decoding

Theoretical analysis of token acceptance conditions in speculative decoding with greedy decoding and tree-based candidates for practical LLM acceleration.

Ax Bertram Taetz, Hugo Albuquerque Cosme da Silva, Gabriele Bleser-Taetz 14d ago

Towards Continual Motion-Language Agents: LoRA Variants for Incremental Motion Understanding and Generation

Methods using LoRA variants for continual learning in motion-language agents handling both motion-to-text and text-to-motion without catastrophic forgetting.

Ax Jan Stenner, Alexander Kilian, Sebastian Peitz, Hermann de Meer 14d ago

Toward an Energy-Optimized Operation of Data Centers Located in Wind Farms Using Reinforcement Learning

RL controller for workload shifting in wind-turbine-integrated data centers with simulation framework for benchmarking energy optimization.

Ax Yousuf Moiz Ali, Jaroslaw E. Prilepsky, Jo\~ao Pedro, Sasipim Srivallapanondh, Antonio Napoli, Sergei K. Turitsyn, Pedro Freire 14d ago

Hybrid Active-Online Learning Framework for Label-Efficient Concept Drift Adaptation in Optical Network Failure Detection

Hybrid active-online learning framework for optical network failure detection adapting to concept drift with margin-based selective labeling.

Ax Marek Polewczyk, Maximilian Schambach, Marco Spinaci, Sam Thelin, Johannes H\"ohne 14d ago

FlexTab: A Flexible Encoder-Decoder Architecture for In-Context Learning Across Diverse Tabular Tasks

Encoder-decoder architecture for in-context learning on tabular data producing target-agnostic row embeddings reusable across diverse downstream tasks.

Ax Haisen Luo, Yiwei Liu, Haoning Wang, Dan Liu, Junxi Yin, Haotian Wang, Lei Zhang, Xiaoyu Tian, Shuaiting Chen, Yuansheng Song, Baoyan Guo, Xiongfei Yan, Bolan Yang, Chengwei Liu, Ming Cui, Jiong Chen 14d ago

DRIFT: Difficulty Routing Self-DIstillation with Rhythm-Gated Exploration and Success BuFfer Training

Self-distillation method for LLM reasoning that routes training by problem difficulty and maintains success buffer for stable improvement without external supervision.

Ax Kuangpu Guo, Qingyan Zheng, Jian Liang, Yongcan Yu, Zilei Wang, Ran He, Tieniu Tan 14d ago

On the Vulnerability of Parameter-Level Defenses to Model Merging

Analysis showing parameter-level defenses against model merging are vulnerable due to small task vector magnitudes enabling reconstruction attacks.

Ax Zheming Fu, Ruizhe He, Wei Shang, Xiaoxiao Ma, Lei Wang, Chang Liu, Siming Fu 14d ago

FlowAWR: Online Adaptive Flow Reinforcement via Advantage-Weighted Rectification

Method for aligning generative flow models via online RL that addresses trajectory likelihood tractability and training-inference inconsistencies.

Ax Pedro Jim\'enez-Gonz\'alez, Miguel C. Soriano, Lucas Lacasa 14d ago

Scalar Representations of Neural Network Training Dynamics

Application of scalar embeddings to analyze neural network training trajectories as temporal networks for understanding optimization dynamics.

Ax Tianyu Wang, Gourav Rattihalli, Aditya Dhakal, Junbo Li, Zhiwei Ren, Dejan Milojicic, Longfei Shangguan 14d ago

Predict, Reuse, and Repair: Accelerating Dynamic Sparse Attention for Long-Context LLM Decoding

Speculate-reuse-repair runtime optimizing dynamic sparse attention for long-context LLM decoding by exploiting temporal locality in block selections.

Ax Lennart Purucker, Andrej Tschalzev, Nick Erickson, Gioia Blayer, David Holzm\"uller, Alan Arazi, Alexander Pfefferle, Mustafa Tajjar, Ga\"el Varoquaux, Frank Hutter 14d ago

Beyond IID: How General Are Tabular Foundation Models, Really?

Evaluation of tabular foundation models on diverse out-of-distribution tasks revealing limitations of current benchmark protocols and model generalization.

Ax Alexis Jacq, Guillaume Couairon, Valentin De Bortoli, Quentin Berthet, Arnaud Doucet, Romuald Elie 14d ago

Diffusion Fine-tuning with Rewarded Moment Matching Distillation

Framework combining diffusion model distillation with RL fine-tuning via Rewarded Moment Matching Distillation to improve generative quality.

Ax Jinda Lu, Kexin Huang, Junkang Wu, Shuo Yang, Jinghan Li, Chiyu Ma, Shaohang Wei, Xiang Wang, Guoyin Wang, Jingren Zhou 14d ago

Experience Augmented Policy Optimization for LLM Reasoning

Method for improving LLM reasoning via reinforcement learning with verifiable rewards by reusing accumulated experience rather than on-policy optimization from scratch.

Ax Ran Canetti, Shafi Goldwasser, Or Zamir 14d ago

Proofs of Ownership for Machine Learning Models

Formal framework for proving ML model ownership through game-theoretic analysis between model owner, thief, and judge.

Ax Liang Wang, Zhaoyang Xi, Zekai Xiang, Heng Meng, Qishan Zhang, Pingyi Zhou, Jin Liu, Litao Chen 14d ago

Arko-T: A Foundation Model for Text-to-Structured 3D Generation

4B-parameter text-to-design model generating executable parametric CAD programs from natural language descriptions for mechanical part design.

Ax Huaqing Zhang, Jingchu Gai, Juno Kim, Bingbin Liu, Andrej Risteski 14d ago

When Does Online Imitation Learning Help in LLM Post-Training? The Role of (Non-)Realizability Beyond Horizon

Research on when online imitation learning improves LLM post-training, showing benefits depend on realizability rather than error accumulation reduction.

Ax Max Fomin, Elad David, Amit LeVi 14d ago

Internal-State Probes Read the Situation, Not the Action: Three Negative Results for Pre-Action Misalignment Monitoring

Study of internal-state probes for monitoring AI agents, finding they read situation context rather than enabling pre-action misalignment detection across model families.

Ax Myung Jun Kim, Maximilian Schambach, Frank Essenberger, Andre Sres, Johannes H\"ohne 14d ago

Exploring Differences Between Tabular Enterprise Data and Public Benchmarks

Analysis of statistical characteristics and performance measurements of enterprise tabular data versus public ML benchmarks for business applications.

Ax Songxin Zhang, Zejian Xie, Zhuoyang Song, Cong lin, Junyu Lu, Jiaxing Zhang, Bingyi Jing 14d ago

HSAP: A Hierachical Sequence-aware Parallelism for Hybrid-Context Generative Models

Proposes HSAP sequence parallelism framework for hybrid-context packed sequences in large language models, fixing cross-contamination in causal attention.