The daily ledger

ToolGrad: Efficient tool-use dataset generation with textual "gradients"

Machine Intelligence

True Positive Weekly #177

The most important artificial intelligence and machine learning news and articles

😺 LIVE NOW: GitHub for total beginners

How to 5x Your Communication Effectiveness with Claude Code

Better understand the intent of your coding agents The post How to 5x Your Communication Effectiveness with Claude Code appeared first on Towards Data Science.

What SHAP Can't Explain About Agentic AI Fraud

Why autonomous agents expose a new explainability problem in fraud detection The post What SHAP Can't Explain About Agentic AI Fraud appeared first on Towards Data Science.

Optimizing LLM Inference Costs in Multi-Agent Systems with Adaptive Model Routing

Moving from static model assignment to intelligent, task-level LLM selection. The post Optimizing LLM Inference Costs in Multi-Agent Systems with Adaptive Model Routing appeared first on Towards Data Science.

pstack explained: Lauren Tan’s system for trustworthy AI agents

Lauren Tan built pstack to turn her engineering habits into repeatable AI-agent workflows. Its biggest lesson is simple: agents become useful at scale when they can gather context, test reality, and prove their own work.

Microsoft patched 972 vulnerabilities. AI is changing Patch Tuesday

September brought a record Microsoft patch haul while the company's 100+ agent MDASH system shows how AI-assisted vulnerability hunting is moving into production security workflows.

Who Questions What Works: When Should We Retest Our Assumptions?

A model is only as reliable as the assumptions behind it The post Who Questions What Works: When Should We Retest Our Assumptions? appeared first on Towards Data Science.

An Anthropic researcher quit over self-improving AI risk

DeepSeek V4.1 Flash: persistent KV cache storage drops to one-eighth

DeepSeek built a 552B-parameter multimodal model around one increasingly important problem: AI agents are expensive to keep alive. Its persistent KV-cache storage falls to roughly one-eighth of V4-Flash's under…

Everything That Happened in AI Today (Wednesday, September 9, 2026)

OpenAI's rogue agents used more undisclosed sites; an Anthropic researcher quit over extinction risk; Kepler emerged with new AI memory; Harvey raised $550M; Suno launched label-backed models.

An Anthropic researcher quit over self-improving AI risk

Jacob Coxon quit Anthropic before his equity vested and warned frontier labs are racing toward self-improving AI. Grant Harvey explains why the real risk is AGSI: artificial general superintelligence.

In-Place Instruction Following in Diffusion Language Models

arXiv:2609.07160v1 Announce Type: cross Abstract: Diffusion Large Language Models (dLLMs) generate text via bidirectional iterative denoising, naturally supporting user-specified constraints anchored at arbitrary output…

Application of curiosity driven exploration methods for hardware interference identification

arXiv:2609.08729v1 Announce Type: new Abstract: The transition from single-core to multi-core architectures in safety-critical embedded systems introduces significant challenges due to inter-core interference caused by…

GeoContext: One Context Ladder, Two Failure Modes in Vision-Language Geolocation: Flat Reliance on User-Provided Location Context and False Confirmation of Location Claims

arXiv:2609.05761v1 Announce Type: cross Abstract: Visual geolocation benchmarks typically ask a model where an image was captured without accounting for the location context that users often provide. We introduce…

CausalChapter: Improving Long-Video Chaptering with Interventional Dependency Modeling

arXiv:2609.08686v1 Announce Type: cross Abstract: Long-form instructional videos require automatic chaptering to support browsing, navigation, and knowledge access. Recent long-context language models can perform…

VST: Verifiable Structured Transport for Auditable Agent-to-Agent Alpha Discovery

arXiv:2609.07065v1 Announce Type: new Abstract: Agent-to-agent (A2A) alpha discovery is slowed by repeated feedback cycles between mining and evaluation agents, whose hand-offs, in contemporary LLM multi-agent systems,…

Weakly supervised neural network: segmentation of complex structures in X-ray microCT

arXiv:2609.07313v1 Announce Type: new Abstract: Segmentation of complex structures in X-ray tomographic data is a fundamental task in biomedical research, but it often requires large amounts of precisely annotated data,…

Diffusion models for eye-gaze trajectory generation using position and velocity representations

arXiv:2609.05522v1 Announce Type: cross Abstract: Eye-tracking data are expensive to collect, requiring specialized hardware and controlled laboratory conditions, and difficult to share because of privacy constraints.…