Weekly Notes 21/2024

May 25, 2024

This week: memories of my grandmother, programming with ChatGPT, monsoon appreciation, Mac gaming, edtech insights and AI links.

For folks not in the know, this is a weekly stream of consciousness zero-agenda “here’s what’s happening” or “here’s what’s interesting” post.

What’s been happening

My grandmother taught me these Ramayanam and Mahabharatham in one slokam when I was a kid, and I think of them fondly often. I thought I’ll dig them up this week, and it’s amazing how much I still remember from all those years ago. She used to tell me that we should recite these before going to bed every day, and then “moksham kittum” or we’ll attain moksha. I have very fond memories of my grandmother, and I hope that my kids will have some of these memories too.

I’ve been using ChatGPT (& Copilot, but less so) as an iterative tool for a lot of programming tasks this week. It’s great for bouncing off ideas, surfacing information, and asking questions like: “What possible security problems can this block of code have?” In some ways, I think of it now as a more advanced (& sometimes unreliable) REPL: I’ve always been a fan of iterative programming through a REPL: figuring out the solution first before polishing it and making it perfect, and if you orient your thinking around it being a REPL, it also makes sense to ask tons of followup questions to get to a solution.

Monsoons have begun in earnest this week. In Malayalam, we call this rain edavappathy (literally the rain that comes in the middle of i?avam). I’m a rain aficionado, and this is my favourite time of the year. This week when we drove down to Kochi, it was raining and took far longer, but we had made our sunroof transparent and it was wonderful to watch the rain falling all around us.

The smell of rain

AAA games have begun to launch on Day 1 on the Mac (App Store link). I hope this trend continues and it’s picked up by more publishers. There are roughly more than ~50 million Macs out there, and that’s roughly about the same as the number of Playstation 5 sales. Even if games are locked to specific CPUs: “you need an M2 processor to play this game” it’s a net win compared to the situation we have now. Just a note that I’m unlikely to play this particular game: I’ve never really gotten into the Assasin’s creed franchise, even though it should check all the boxes for me.

I’ve been reading Salman Khan’s Brave New Worlds, and while portions of it read like an advertisement for Khanmigo, it also contains a lot of lessons for anybody building an education technology startup today. It hits pretty close to home as I was the founder of an edtech previously, and the primary issue in education is what I like to call Bloom’s “not enough teachers” problem. AI provides the first inkling of how technology can help here: by being an always patient, eternally available 1-1 tutor that can employ techniques like socratic questioning and mastery learning. With AI, we can also develop entirely new ways to learn content: Sal mentions a couple: imagine writing an essay on a book and being able to talk directly to one of the characters and ask them questions, or learn history by asking historical figures like Gandhi what they would have done with India’s current problems. It’s got a bunch of nice nuggets of information, and is a recommended read if you are interested in edtech. For a teaser of what the book contains, watch the video instead.

I tried using the Msty app with the Llama 3 model this week, and while it’s really cool to run an LLM locally, the results are less than impressive. This is where the technology is trending though: soon you’ll be able to have GPT-4-level AIs running on device. What possibilities will that unlock?
Loved this exploration of the insides of an AI, and how it correlates various features within it. Isn’t this very similar to our human brain?
Tutory seems to be like a more available Khanmigo.