Trace and evaluate any AI/LLM app in 3 lines of code! - Initialize an eval with LangWatch (line 3) - Decorate the LLM workflow method (line 5) - Log the evaluation (line 11) You can keep using Pandas and your other favorite tools. 100% open-source!

RIP Perplexity. You don’t need to spend $20/month anymore. You can now turn any LLM like ChatGPT, DeepSeek, Gemini, or Qwen into a 24/7 research agent. Here’s the exact mega prompt I use to automate all research for free:

you can figure out anything with printf

wasting tokens on bullshit

lets fucking go dude

okay computer, put on "fly like a g6" but make it nightcore

it's so depressing. i catch one night of good sleep and then i'll figure something out that i was stuck 3-4h on

paradoxically if i sleep it means i'll have more time to actually sleep

how your email finds me

there's just something about this that is extremely demonic and i think its about the disrespect i would be giving my child to give him something that wasn't an actual real computer

The most important question to ask yourself: are you having fun?

I got seriously lucky early as a kid, for all the care that my parents gave me when I was in middleschool. I got tutored with math at home, my mom spent a lot of time teaching me to read. They got us books. For the longest time I thought that this was just the default

I didn't realize it at the time, but my dad taught me how to think by making me do euclidian geometry proofs. I thought he was just teaching me math, but now I realize it was pretty purposeful

I know that this comes from a place of privilege but man if you got to university in the first place you can learn anything. Like fundamentally, you can actually grind out numpy backpropagation manually every day until it just sticks. This is totally possilbe

people just dont they're lazy

the custom framework to avoid being in overfit programming space ^^ it works astoundingly well

you should set up a small validation set of like ~50-100 short examples where you vibecheck for edgecases/failures before you build rubric criteria this way, you have some signal for what rubric criteria is better via your "judgement set as eval" (for consistent filtering)

me: it's really important to keep your cost of living and lifestyle low, even as your means increases also me: i'm buying my 6 month old 4 laptops so i can teach him how to use arch linux

because he keeps on reaching for mine, i'm buying my 6 month old a laptop you're screen avoiding, i'm screen maxxing

i'm getting him a thinkpad probably. what's the meta for ultra cheap laptops? i'm going to buy 4 and put arch linux tty one them and program it to make sounds and shit