Join 📚 Josh Beckman's Highlights

A batch of the best highlights from what Josh's read, .

Data caching is probably the most effective and most essential layer you should be working on, before moving up to more advanced caching.

Production Ready GraphQL

Marc-Andre Giroux

The other thing about chat was when we had these instruct models. The task of “complete this text,  but in a nice or helpful way” is a pretty poorly defined task. That task is both confusing for the  model and for the human who's supposed to do the data labeling. Whereas for chat, people had an intuitive sense of what a helpful robot should be like. So it was just much easier for people to  get an idea of what the model was supposed to do. As a result, the model had a much more coherent personality and it was much easier to get pretty sensible behavior robustly.

John Schulman (OpenAI Cofounder) - Reasoning, RLHF, & Plan for 2027 AGI

Dwarkesh Patel

Taxes *don’t have to be about revenue*. You can institute a wealth tax because inequality is bad, just like you can tax cigarettes because smoking kills. Inequality, as a social ill, is a problem to solve, not just a source of revenue.

I Read Capital in the Twenty-First Century by Thomas Piketty on August 21, 2020

Tom MacWright

...catch up on these, and many more highlights