AI Research

True Positive Weekly #181

The most important artificial intelligence and machine learning news and articles

😸 WATCH: Can AI Help Cure Genetic Disease?

Everyone Is Selling AI at You — Here’s How to Keep Your Judgement

Three moves to decode the hype The post Everyone Is Selling AI at You — Here’s How to Keep Your Judgement appeared first on Towards Data Science.

Google Wants a Bigger Role in How Work Gets Done

Google wants Gemini to take on more responsibility at work. The practical test is how much attention teams save after reviewing and correcting its output.

AI Agents Beat PyTorch: Writing Faster CUDA Kernels

AI agents can now write CUDA kernels that outperform PyTorch—but proving those speedups are real is the harder problem. I put them to the test on an NVIDIA DGX Spark and found that benchmark design matters just as much…

Your Model's MSE Is Lying to You III: Time Series Diffusion

A diffusion head for probabilistic time series forecasting: why a forecast can get the mean and the variance right and still describe a future that never happens, and how to fix it. The post Your Model's MSE Is Lying to…

Why Meta, DeepMind and the U.S. Government Are Backing Virtual Cells

Meta, Google DeepMind and U.S. agencies are backing Biohub’s effort to build the experimental data AI needs to predict cellular behavior. The promise is faster discovery; the test is whether those predictions hold up in…

Haiku 5.5 Could Be Claude’s Most Practical Launch Yet

Claude’s smaller model has a timely opening: agents need affordable intelligence they can call repeatedly. Haiku 5.5 delivers lower rates, but reasoning effort, growing context, and completion quality determine the real…

AI Engineers, Applied Scientists, Forward Deployed Engineers: Let's Unpack.

An exploration of the new data science roles emerging after the Large Language Model explosion. The post AI Engineers, Applied Scientists, Forward Deployed Engineers: Let's Unpack. appeared first on Towards Data Science.

Google Wants SynthID to Become the Web’s AI Watermark Checker

Google’s SynthID Detector can now check AI watermarks from OpenAI, NVIDIA, Kakao, and Google itself. The expansion could make AI content verification easier, but an undetected watermark does not prove authenticity. As…

😺 ChatGPT's GPT-6 now builds calculators

OpenAI’s GPT-6 Turns ChatGPT Answers Into Interactive Tools

GPT-6’s Intelligent UI lets ChatGPT assemble interactive tools as it responds. The rollout could change how people use AI and whether they open a separate app to complete a task.

Estimating Model-Level Membership Inference Vulnerability Without Reference Models

arXiv:2510.19773v2 Announce Type: replace Abstract: Membership inference attacks (MIAs) have emerged as the standard tool for evaluating the privacy risks of AI models. However, state-of-the-art attacks require training…

The Silhouette Operator: Identifiability of Low-Rank Measures from One-Dimensional Projections

arXiv:2610.09687v1 Announce Type: cross Abstract: Structured recovery phenomena, such as restricted isometry properties in compressed sensing, have shown that high-dimensional objects can often be reconstructed from…

Global Exponential Convergence of Two-Layer Linear Network Training

arXiv:2610.09356v1 Announce Type: new Abstract: We prove global exponential (linear) convergence with an explicit rate in the rich scaling for wide two-layer linear networks trained with smooth Polyak-Lojasiewicz…

The Dichotomy Between Pattern Recognition and Step-by-Step Reasoning

arXiv:2610.09186v1 Announce Type: new Abstract: We argue that pattern recognition and step-by-step reasoning are two ends of a spectrum. A large language model (LLM) learns to reason step-by-step when data is structured…

Stream-Based Active Learning with Cooperative Neural Networks for Data-Efficient Partial Inverse Design: An Automotive Glass Run Channel Case Study

arXiv:2610.09848v1 Announce Type: new Abstract: Inverse design in engineering often runs into a simple problem. Each labeled training sample must be produced through expensive simulation, so building a large dataset is…

Residual Learning in Empirical Asset Pricing

arXiv:2610.09613v1 Announce Type: cross Abstract: Shallow models are special cases of deep models, and deep models theoretically have the potential to outperform the shallow ones. However, the existing empirical asset…

FAR: Failure-Aware Retry for Test-Time Recovery and Continual Policy Improvement

arXiv:2607.01111v2 Announce Type: replace-cross Abstract: Robot policies inevitably encounter failures when deployed in real environments. Naive retries often repeat the same mistakes, while many existing recovery…

When does a network's training history predict its future learning better than its current state? Evidence from a response probe and a forecasting screen

arXiv:2610.09621v1 Announce Type: new Abstract: Networks that behave alike now can still learn differently when training continues. Work on loss of plasticity and critical periods shows that the path to a state shapes…