The Week AI Blinked: the limits of smart

Hey there Ai5ers,

From identifying AI's limitations and confusion to insights from leaders in the industry, this weeks newsletter is a whirlwind.

Happy reading!

Today's Ai5:

🧐 Tackling hallucinations: MIT spinout teaches AI to admit when clueless
😵 Anthropic's AI generated blog dies an early death
🤓 Still no AI powered Siri from Apple
💰 Exclusive: OpenAI taps Google in unprecedented cloud deal
🤖 How To Make AI Faster and Smarter With Physics

Prompt of the Day 🎨

Prompt with your pups: "Attached are the images of my puppies. Generate dramatic, photo-realistic scene of the two of them, running on a dirt road in a rural landscape as a massive tornado swirls behind them. The sky is filled with dark, stormy clouds and flying debris. The lighting is moody and cinematic, with the tornado looming in the background, capturing a sense of urgency and motion. Last, in [white Lobster font] title at the top of the image, "Where are we going? Anywhere but here!"

Tackling hallucinations: MIT spinout teaches AI to admit when clueless 🧐

AI hallucinations are becoming more dangerous as models are increasingly trusted to surface information and make critical decisions.

We’ve all got that know-it-all friend that can’t admit when they don’t know something, or resorts to giving dodgy advice based on something they’ve read online. Hallucinations by AI models are like that friend, but this one could be in charge of creating your cancer treatment plan.

AI systems typically display overconfidence. Themis’ Capsa platform acts as a reality check for AI, helping models recognize when they’re venturing into guesswork rather than certainty.

Full article by Ryan Daws

Anthropic's AI-generated blog dies an early death 😵

Sometime over the weekend, the Claude Explains blog disappeared — along with its initial few posts.

Claude Explains didn’t get the rosiest reception on social media, in part due to the lack of transparency about which copy was AI-generated. Some users pointed out it looked a lot like an attempt to automate content marketing, an ad tactic that relies on generating content on popular topics to serve as a funnel for potential customers.

Anthropic might’ve also grown wary of implying Claude performs better at writing tasks than is actually the case.

Full Article by Kyle Wiggers

Still no AI-powered Siri from Apple 🤓

At this year’s Worldwide Developer Conference (WWDC 25), Apple announced a slew of updates to its operating systems, services, and software, including a new look it dubbed “Liquid Glass” and a rebranded naming convention. Apple was notably quiet on one highly anticipated product: a more personalized, AI-powered Siri, which it first introduced at last year’s conference.

While Bloomberg reported that the in-development version of the more personalized Siri was functional, it was not consistently working properly. The report said its quality issues meant Siri only performed as it should two-thirds of the time, making it not viable to ship.

Full article by Sarah Perez

Exclusive: OpenAI taps Google in unprecedented cloud deal 💰

OpenAI plans to add Google cloud service to meet its growing needs for computing capacity, marking a surprising collaboration between two prominent competitors in the artificial intelligence sector.

It underscores how massive computing demands to train and deploy AI models are reshaping the competitive dynamics in AI, and marks OpenAI's latest move to diversify its compute sources beyond its major supporter Microsoft.

Full article by Kenrick Cai and Krystal Hu

How to Make AI Faster and Smarter With Physics 🤖

Rose Yu has drawn on the principles of fluid dynamics to improve deep learning systems that predict traffic, model the climate, and stabilize drones during flight.

The work has not only introduced novel techniques for building and training these systems, but it’s also allowed her to make progress on several real-world applications. She has drawn on principles of fluid dynamics to improve traffic predictions, sped up simulations of turbulence to enhance our understanding of hurricanes, and devised tools that helped predict the spread of Covid-19.

As I picture it, an AI Scientist could relieve researchers of some of the drudgery while letting people handle the creative aspects of science.

Full article by Steve Madis