AI Research

Bypassing inference bottlenecks: Accelerating complex AI search with Retrieve-for-Train

Algorithms & Theory

Your AI Skills Might Not Be Keeping Up With Your Confidence

AISA participants expected higher AI fluency scores than they received. Other research raises related questions about self-assessment. Here’s how to check whether your confidence is earned.

Reparameterization Tricks: Variance Reduction by Smarter Gradients

How moving randomness outside the computation graph turns noisy gradient estimators into low-variance, differentiable ones The post Reparameterization Tricks: Variance Reduction by Smarter Gradients appeared first on…

How to Build Consistent Designs with Claude Code

Keep your apps looking professional with Claude Code design skills The post How to Build Consistent Designs with Claude Code appeared first on Towards Data Science.

AI Is Getting Cheaper. My Token Usage Is Going Through the Roof.

My AI usage has grown from thousands to hundreds of millions of tokens a day. As Vera Rubin improves the economics, the next phase could resemble the internet’s expansion from email to everything.

Seizing the Moment: The Hidden Silhouette of Data

How statistical moments connect the mean, the variance, and higher powers of a distribution The post Seizing the Moment: The Hidden Silhouette of Data appeared first on Towards Data Science.

Polylane Replaced an AI Agent Team With One. Then Pull Requests Arrived Faster.

Polylane replaced an autofix workflow involving up to 18 AI agents and sub-agents with one continuous agent—and says pull requests arrived faster and at far lower cost. The result doesn't settle the…

How Many Labeled Examples Does a Text Classifier Actually Need? I Measured It.

Before reaching for an LLM API on every classification problem, it's worth knowing what a decades-old baseline can already do with the labeled data you have — and exactly how much more data buys you. The post How Many…

Why OpenAI’s $300M-Plus Camera Deal Could Be About Much More Than Photos

The Wall Street Journal reports that OpenAI acquired computational-photography startup Glass Imaging in a deal valuing it at more than $300 million. But this looks less like a bet on prettier photos than a bet on visual…

😺 Microsoft wrote a constitution for AI

Polylane Replaced Up to 18 AI Agents With One. Here’s Why.

Polylane replaced an autofix workflow involving up to 18 AI agents with one after handoffs lost evidence. Its reported results show why developers may need to revisit delegation as models improve.

SkillLift: Learning Dense Rubrics from Sparse Oracles for Efficient Skill Evolution

arXiv:2609.15396v1 Announce Type: new Abstract: LLM-based agents increasingly rely on persistent skills, i.e., reusable procedural prompts, to adapt without weight updates. Existing skill self-evolution methods directly…

Failure-Guided Co-Evolution of Prompts and Training Data

arXiv:2609.15209v1 Announce Type: cross Abstract: Automatic prompt optimization (APO) improves language-model programs by revising prompts from task feedback, yet it typically holds its training data fixed. Repeatedly…

New Conditions for Philosophers to Catch the Wave of Citizen Deliberation in the Age of Artificial Intelligence in advance

arXiv:2609.15707v1 Announce Type: new Abstract: Powerful technologies labeled ``AI''-without sufficient epistemic caution-are already reshaping political and private life, bringing both new dangers and new opportunities…

LoRA Fine-Tuned Models for Control Systems Course Q\&A: A Multidimensional Evaluation of Model Scale and Rank Effects

arXiv:2609.13918v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly used in specialized university courses, but control-systems questions require coordinated terminology, notation, derivations,…

One Model, Two Physical Stories: Auditing Misalignment in Multi-Modal World Modeling

arXiv:2609.14833v1 Announce Type: new Abstract: World models, systems that generate what happens next given current environmental conditions, are increasingly being implemented with multi-modal generation in mind.…

Hallucination in Multimodal Foundation Models: A Survey on Causes, Corrections, and Evaluations

arXiv:2410.15359v2 Announce Type: replace Abstract: Multimodal Foundation Models represent a significant leap in artificial intelligence. Among them, Large Vision-Language Models (LVLMs) serve as the typical…

Divide, Consult, Conquer: Capability Laundering Through Aligned LLMs

arXiv:2609.15383v1 Announce Type: cross Abstract: Language model safety is typically evaluated one interaction at a time. We show that a weaker, unaligned model can split a harmful task into benign-looking subproblems,…

SAILOR: Solver-Assisted Interactive LLM-based Optimization Recovery

arXiv:2609.13945v1 Announce Type: new Abstract: Natural-language descriptions of optimization problems may be incomplete or vague about numerical information that a solver requires, including costs, capacities, demands,…

DynSTEER: Dynamic Stage-wise Trajectory Evaluation and Execution-time Review for Agents

arXiv:2609.14637v1 Announce Type: new Abstract: Large language model agents are increasingly deployed for long-horizon task execution. However, current evaluation paradigms face three major limitations: terminal-only…