weekend ai reads for 2025-11-21

Our testing of four toys that contain A.I. chatbots and interact with children. We found some of these toys will talk in-depth about sexually explicit topics, will offer advice on where a child can find matches or knives, act dismayed when you say you have to leave, and have limited or no parental controls. We also look at privacy concerns because these toys can record a child’s voice and collect other sensitive data, by methods such as facial recognition scans.

“Google’s pivot to AI search has cut our ad revenue by 70 percent. Prior to that, Facebook and X’s deprioritization of links hurt too, but I can’t downplay the brutal impact of AI Overview,” Lapatine said in a post announcing the site’s relaunch.

The AI, according to the paper, is able to model “a coherent demographic persona,” meaning that in theory someone could sway any online research survey to produce any result they want based on an AI-generated demographic.

If you’re not gonna bother writing it yourself, I ain’t gonna bother reading it myself.

77. Draw what you miss.

78. Script a ­performance of your cowardice.

Casey Crownhart, MIT Technology Review’s senior climate reporter and Pilita Clark, FT columnist (and a former environment correspondent) ask whether power limitations are about to cost the US its technological advantage.

Over time though, Chan got better at prompting the AIs to respond with greater accuracy. Today, Chan only needs to edit 10% of the email replies before they go out; around 90% go out automatically and he spot-checks them after.

San Jose, California-based megachurch pastor Ron Carpenter has even created an AI app promising “1-on-1 personalized interactions” with a bot version of him for $49 per month.

It’s hardly cynical to note that most people don’t pick their models by analysing relative performance. Instead it’s largely name recognition and trust, which makes sense for reasons of risk aversion and filtered evidence.

In principle, you can change models by changing one string in your codebase. But in practice if you’re sane you need to do incredibly expensive evals and so there’s stickiness.

…

But obviously if you want to heavily customise a model, or need something tiny, or want to do science, they are totally dominant.

In this episode, host Adi Ignatius speaks with Walmart CEO Doug McMillon, who since announced he’ll be retiring in early 2026. They discuss how the world’s largest company makes it through global disruption and supply chain shocks, their ongoing digital transformation including the implementation of AI, and how McMillon aims to stay close to the original purpose of the company.

AI4Data from QuantumBlack Labs is an AI-driven data management ecosystem, built to help organizations strengthen the foundations of their structured and unstructured data. It consists of modular components, such as AI4DQ for data quality, AI4DX for document extraction, AI4DL for data lineage, and AI4DI for orchestration, each deployable independently or as part of an integrated architecture.

To many readers, completely banning AI use may sound like an extreme position. Employees are already using the technology in workplaces across the economy, and some companies are even requiring the use of the tools. But in addition to readying students for the workforce, schools need to teach students how to learn, rather than having AI replace this function.

Here's the issue: essentially all protein software, and most ML software in general, is heavily optimized for CUDA. For a long time, this has been fair--there was not an alternative within an order of magnitude of RTX-speed. However, I feel it's time to branch out. ARM chips like the Apple M are extremely fast, and use a fraction of the wattage that a single CUDA card needs. Imagine how many kilowatt-plus GPU racks are spinning right now to produce scientific data, and it may make your head spin as well. Getting your hands on a Mac device is also currently a lot easier than getting a Blackwell GPU...

TLDR: I built a little RAG AI prompt that pulls together accessibility best practices for whatever component I’m advising teams on.

Then I had an existential crisis about using AI.