Nah. Weak take. There are more than enough children’s books that have been published and continue to be published that kids love. But if you don’t turn them on to the love of books and they default to staring at screens, then yeah, kids aren’t going to read.

With all the AI code generators reducing the cost of shipping web apps the value of good product design is going to go way up. Long PMs and designers! 😄

Breakthrough with Medthrive.ai yesterday. Turns out just adding more context to the prompt works.

LLMs seem to be really bad at randomization. Using Claude Sonnet to create multiple choice questions with 5 options and ~70% of the time the correct answer is the first. I wonder if we can prompt to better randomization or whether I’ll need some post processing…

Mental model of LLMs right now: Not actually that smart, but good at understanding when a response is incorrect, identifying potential solutions and looping until they arrive at a reasonable answer. Human experts are smarter but LLMs are better at quickly iterating on solutions

Tested Gemini on converting images of freehand notes to text earlier and it blew open ai, mistral and even Claude out of the water. Not just accurate but crazy fast. And no one seems to have noticed?

Posted a silent recording to the OpenAI whisper model and it sent me this website lmao zeoranger.co.uk

The insanity of creating and posting this aside, it’s hard to imagine an image that says more about the US in 2025 than this.

AGI is cool, but if you want to make money you need to tap into existing profit pools. Makes sense that OpenAI should start pushing aggressively into enterprise. Interesting to think what the AI era’s office suite looks like.

A new era begins for Hinge Health. Over a decade ago, we set out to reimagine healthcare—today, we celebrate going public as $HNGE on the NYSE 🏛. Thank you to our members, partners, clients & team who made this day possible.

Examples of LLMs generating UI on the fly getting increasingly common. Squint and you could see how this enables LLMs as the OS. Unclear whether that’s actually desirable though. Pretty compelling for exploring data, however.