← back

What I Learned This Week

4 min read ·

A little bit about how vtables, that make dynamic dispatch and polymorphism possible, through a video by low-level learning

There is a series of photos of people playing chess on roller coasters because of this xkcd post

We can make floaty things

Open-source machine learning models may be a thing, at least for some on-device use cases. There are also some great lists of resources on LLMs and the AI ecosystem in general

Every time I learn a little more about encryption it feels like more magic, like the way you can safely exchange keys in an unsafe channel. This week I learned that you can check if something is on a list without knowing the members of the list. Also, while learning about federated learning, homomorphic encryption showed up which lets you use encrypted data as if it wasn’t.

I’m listening to Measure What Matters and the Larry Page intro and Andy Grove intel stories got me sold on OKRs. I’m going to try it out on Mind and my personal life, maybe even publish my results. There is also a passage on how the industrialist (think Henry Ford) way of top-down low-level management doesn’t work for knowledge workers and a new way was needed, that matches perfectly with what I’ve been discussing with Gustavo on how to measure productivity. The idea is that on a factory line, you can easily map an input to output, the more or faster mechanic movement makes it deliver more, and slacking off hurts the company. Programming or other creative work requires some wandering as you can’t have clever ideas on demand. Another thought that got turned into words is that it’s better to have a flatter org chart, the “leaf” nodes have more responsibility and the company feels more like a collective.

Nicole Ruiz, a former VC at Compound

About Compound, it’s VC that focuses on business with a thesis behind them. They have a public series on their thesis and I found it an awesome resource for thinking of new businesses and products. Reading about Science Driven Mental Health got me thinking about how could we get it into Mind, maybe an integration with pharma(?). What if you could manage your patient-therapist-psychiatrist relationship via an app integrated with drug delivery/development?

People are creating tinder profiles on google docs and notion. It’s pretty clear that all dating platforms lack something, you can verify it by the huge dissatisfaction surrounding it. In my opinion, the problem isn’t the lack of information about the people there, the candidates, but excessive access and choice that you have.

There is a search engine start-up that shares the revenue of its ads and it’s made by the same guys at ahrefs (they don’t like the cloud btw, and have a shit ton of GPUs)

Corecursion is a thing though it doesn’t seem very useful. It has a base case, the same way as recursion, but differently from recursion, that goes toward the base case, it starts off from that and generates different cases. (I clearly don’t completely get it, need to explore further)

Afterthoughts

I started with this idea at the start of the week with the expectation of posting it on the weekend, and my experience so far has been really great. I noticed I pushed myself to learn more and separating topics into context-free paragraphs made have to think about how each piece of information led to another. From now on I’m going to explore more this idea, a written weekly learning review that gets made on the go, and possibly add a section at the end of my normal blog posts just for that. I could keep it private, but posting them makes me think more about how to explain it so that it’s valuable to other people, also the pressure of public accountability helps on keeping plans going.