I've been asking Claude to quiz me to keep me sharp

This month

A quick housekeeping note: I'll now be sending out the Broadcast monthly instead of weekly. Each issue will be a bit meatier and will have sharper curation and editing.

Things are heating up at Cleverest HQ, so I've had to take a little break from filming for YouTube. But I'm happy to say I'm back again! I never thought I'd be filming content in my car. But there I was, holding up my phone to my face in the driver's seat. It was giving very heavy influencer vibes.

I was doing all this to demonstrate a workflow using Claude's interactive voice mode to triage your email hands-free on your commute (or on a run, at the grocery store... while gardening).

Instead of testing everything out and then filming a polished demo of the workflow, I decided to film the scrappy test version. What you see in the video is the very first time I tested out the triage process with Claude end-to-end.

And scrappy it was indeed! Claude made a really interesting mistake during the test where it deviated from my instructions. It turned into a chat about the difference between rules and guidelines.

Watch it here →

What I'm learning from

I Vibe Coded a Security Risk — Katie Parrott / Every. Parrott writes about relying too much on AI coding assistants as a non-engineer. On the one hand, vibe coding (building software by describing what you want to an AI) is freeing for those of us who are not software engineers. It is reducing the gap between technical and non-technical. On the other, as Parrott says, "The appearance of 'done-ness', to the untrained eye, becomes evidence of done-ness." Confronting my own limits and not leaning on AI to fill my skills gaps has also been an important journey for me.

Claude Fable, Claude Tag, and Anthropic's Culture — Cat Wu & Thariq Shihipar / AI Engineer. Two people from Anthropic's Claude Code team on how they cut Claude's system prompt (the standing instructions that define how it behaves) by 80% for the newest models, trusting the model's judgement over rules and examples. While this made me consider cutting my own instructions, rules, and examples for Claude, my experience has shown that examples and rules can go a long way if you have a very specific result in your head. But it's true that Fable (the newest Claude model) can make very impressive judgement calls and occasionally come up with interesting ideas. Stay for their intro to Claude Tag too: this is the first multiplayer product to work with Claude as a team.

Using AI to Increase Your Intelligence & Enrich Humanity — Dr. Fei-Fei Li / Huberman Lab. Fei-Fei Li, Stanford professor of computer science and often called "the godmother of AI", talks about the AI of the past, present, and future. Among the insights she shares are the similarities between how humans and AI learn. The key difference is that humans can learn new things with relatively few exposures, and can abstract and generalize from them. AI, on the other hand, often needs massive amounts of data. And although AI is trained on all of modern human civilization on the internet, there are still many things, like thoughts, ideas, intuition, and lived experience, that only live inside humans, and thus cannot be accessed by AI. Both of these things combine to explain where AI still faces limits to mimicking real human intelligence. She also highlights how to use AI to increase agency and creativity, and how we should all approach AI in a way that ensures a future where we will be amplified, not replaced.

Try this month

Let your AI quiz you

I've had Claude quizzing me (almost) daily on what I've been reading, listening to, and working on. It assigns XP (a running score, like in a video game) based on how well I answer, and if I'm struggling with something, it resurfaces that topic a few days later. The principle behind it is active recall: pulling something out of memory strengthens it — rereading only gives you the illusion of deeper understanding.

You don't need any setup to try this. At the end of your day, give your favorite AI assistant the things you read or worked on — an article, a book chapter, your meeting notes — and ask:

Quiz me on this. Three questions, one at a time, spread across everything I gave you. No multiple choice, no yes-or-no: make me explain things from memory, and make at least one question "why does this matter" or "how would you use this". Grade each answer 10, 5, or 0. A 10 needs the specific names, numbers or mechanism from the source; the gist alone is a 5; wrong or "I don't know" is 0. Only credit what I said, not what I could have meant. For the why/how question, grade my reasoning, not whether I agree with the source. After each answer, tell me what I missed, then ask the next one. At the end, give me my total and the topic I was weakest on.

If you want to keep score over time, your performance needs somewhere to live between chats. Mine lives in a small log file that Claude reads and updates itself. To do this you'll need an assistant that can edit files on your computer or in the cloud, like Claude Cowork or Claude Code. Add this line to the instructions above:

Keep notes on my strong and weak areas plus my score in quiz-log. Dock it 10% for every day since the last entry, add today's points at the end, and update the file with a summary of today's quiz and my current score.

Then save the instructions as a skill and give Claude access to a file called quiz-log.

Slow note

I was solo parenting last week and of course the universe conspired to give my 2-year-old a cold at the same time. It was frustrating because I had been excited about working on my upcoming tutorial and helping clients with post-summer projects. But with a grumpy toddler who couldn't breathe through her nose at home, I had no choice but to slow down. And it ended up being kind of cozy. My daughter is really into Daniel Tiger audio stories and we found the perfect one for last week: Quiet time at school. It teaches techniques to find rest during the day. The song they sing was our mantra for the week: "Close your eyes, snuggle, or take a deep breath. Do what helps you rest."

Upcoming courses