Cedar @Cedar@schelling.pt

@niplav gib substack critique pls

@redacted yup this could defeat my argument. But now supposed that I argue back that the adult human is better at shape rotating because social learning?

Now I'm all tired and lazy and would rather take a nap than review my distributions.

I made anki deck for reviewing probability distributions, including all the essentials:

- PDF
- PMF
- Mean
- Variance
- explanation. https://github.com/renxida/probability-distributions-anki

Edit the python script if you would like to e.g. add more distributions, scrape distribution related data from elsewhere

Human see human do.

Monkey rotate shape in head.

You're not that much smarter than a chimp. Your advantage is your ability to connect with people. Don't search the strat space yourself, use imitation.

(Source: The Secret of Our Success, Joseph Henrich; reproduced from https://lnkd.in/eJCKFC8q)
---
RT @shreyas
Failure is the absolute worst way to learn something and sadly so many people spend their entire lives under the illusion that failure is the best way to learn anything.
https://twitter.com/shreyas/status/1624965994182766598

TIL about Klong and APL. They are apex arrays processors and in a world where things don't get trapped in local minima as much, GPT3 could have been trained on code written in APL.

prime.html
https://analyzethedatanotthedrivel.org/2018/03/31/NumPy-another-iverson-ghost/

PS: Thanks @niplav Your page sent me down this rabbithole

I guess stop doubting that I'm smart & capable is a step forward. But we'll see.

Late night anxiety loop things:

I have friend who is trying to convince me to quit PhD & go industry.

What if my friend is dramatically overestimating the importance of intelligence & ability at getting a high paying job, and underestimating credentials and connections??!!

What if I quit the PhD and I get nothing for a long long time because I have no credentials to flash and no connections to ply?!

By "roughly", I mean, "rounds to with 1 significant digit"

Or roughly 100 sqcm

this works nicely because 1m = roughly sqrt(10) ft, and 1m^2 (roughly) = 10 ft^2

TIL: a dollar bill is roughly 0.1 sqft
---
RT @AlexKrusz
@ruthhook_ The average California lot size is 8,327 square feet. One dollar bill is 0.111 square feet. So $74,824 of dollar bills will cover one average lot. So the answer is about $75k.
https://twitter.com/AlexKrusz/status/1624329338354307072

@niplav heh ty

Can't wait to try it on a makefile that doesn't compile. Don't think it's gonna work yet but as model capabilities rise maybe crazy things might happen

Oh and also, the fancy name is important.

VERY important.

It's really not that hard at all. https://platform.openai.com/examples/default-fix-python-bugs

OpenAI already has an example, and all I had to do is add some python code to read the file & run it to generate the errors.

I guess this sortta shows how trivial it is to start an "AI Startup". You just call OpenAI's api and write some UI / UX / frame code, and voila!

I made it

https://github.com/renxida/maktlo

I want to make a tool that takes:
1. The path of a small code file
2. A command to compile the code
3. Some natural language context
And feeds compiler errors to GPT3, and ask for a diff on how to change the file, until it compiles.