Just have to say, I’m enjoying playing with llm so much. Being able to easily play with this technology at the command line feels like magic. Thanks for the awesome work!
I've contracted an ex Azure DNS team member to write up articles about DNS [1] and published it for free. I considered my DNS knowledge okay, but I learned something every article he wrote.
If you want to be better at DNS than >99% of your colleagues for the rest of your career, then invest a single day in reading those.
I'm sure there's a lot of good material around, but here are some links that are conceptually very close to the linked Autodidax. (Disclaimer: I wrote Autodidax and some of these other materials.)
There's Autodidact [0], a predecessor to Autodidax, which was a simplified implementation of the original Autograd [1]. It focuses on reverse-mode autodiff, not building an open-ended transformation system like Autodidax. It's also pretty close to the content in these lecture slides [2] and this talk [3]. But the autodiff in Autodidax is more sophisticated and reflects clearer thinking. In particular, Autodidax shows how to implement forward- and reverse-modes using only one set of linearization rules (like in [4]).
There's an even smaller and more recent variant [5], a single ~100 line file for reverse-mode AD on top of NumPy, which was live-coded during a lecture. There's no explanatory material to go with it though.
Heh, decent timing. My last CBT session ended with me realising one of my issues may be a lasting effect of bullying.
Essentially I was trying to find out why I procrastinate, beyond "it's just the ADHD lol"
Boiled it down and down and down until we hit the core of the issue and realised I don't want to poke my head up and ship code/sites/ideas/etc because I believe I'll be bullied for it. In school doing anything that lifted you up (e.g. good grades) made you a target.
The realisation was that that will probably not to happen now. If I do something exceptional at 34 I probably wont be bullied by my peers (or heck, even if I actually was bullied for it.. who cares lol, my peers don't understand the market I work in)
I'd never reevaluated the internalised rule "If I excel, I will be bullied" until the other day.
--
On a slightly more positive spin the whole thing made me quite sensitive to deception. If someone is trying to deceive or manipulate me my subconscious might as well be flashing up a Metal Gear Solid exclaim noise for how obvious it seems.
Not really, they don't often advertise that you should attack their customers directly. The closest I can remember was the LifeLock guy putting his social security number up publicly.
Otherwise, they prefer you hit test or personal accounts rather than paying customers...
I can't recommend enough this "backpropagation from scratch" from heremy howard (fastai) where he implementes backpropagation from scratch using python.
Yes, it should still be! The motion of the planets in the sky is relatively small day-over-day.
A good way to verify yourself would be to use a tool like Stellarium Web [1] and set your location and set the time to tonight at say, 3am (the planets become more visible as you get nearer to dawn tomorrow). You could even change the time from say 6/25 at 3am to 6/24 at 3am to see just how much / little it changes night-over-night.
I just learned a new programming language yesterday in an hour: Svelte. It's just a tiny superset of JS/HTML/CSS, but I learned it and have already produced working web application faster than with Vue / React, and it's a perfect hobby language for me to play with, and a perfect example of why new languages can still give something extra in return. I just link to the tutorial, because it's so well done:
For what it's worth, you could watch how quickly the confidence intervals converge as you sample the data, to see if it's worth continuing or if the variance is too high and whether you'd have to check thousands of pages by hand:
from scipy.stats import binomtest
chance_of_dead_page = binomtest(landing_page_counter["dead"], landing_page_counter["total"]).proportion_ci(confidence_level=0.90)
print(f'Chance of a dead but existing landing page (90% Confidence Interval):{chance_of_dead_page.low * 100:.2f}% to {chance_of_dead_page.high * 100:.2f}%')
One important aspect of leetcode/coding contest problems is that they have a input size constraint and a time limit constraint.
You can use the two to figure out what is the time complexity for a solution that would work. This simplifies the search for a solution by quite a bit. Here's a blog post about this idea (going from the input constraint to the possible algorithm): https://www.infoarena.ro/blog/numbers-everyone-should-know
Other than that, understanding a set of frequent data structures and algorithms helps a ton. Here's a short course from stanford on preparing for coding contests http://web.stanford.edu/class/cs97si/
I tend to read the abstract, figures and captions, conclusion. Then i know if it’s worth really reading it, at which point i print it out and break out a pen.
if it’s really REALLY worth reading, ipython usually comes out to play, too
https://hareeshganesan.com/2024/07/14/baby-calendar