Pinned Tweet
Finished an article summarizing some tips I learned to write faster pandas code! Notebook included!
[buymeacoffee.com/entropychase/f…]
For me, this led to big speed improvements, esp for tasks that require row by row iteration!
[buymeacoffee.com/entropychase/f…]
For me, this led to big speed improvements, esp for tasks that require row by row iteration!
7
12
69
13,515
I need fashion advice: I dislike half windsor tie knots bc they always look wonky and off balance, but Im wondering if skinny guys can pull off a full windsor. The knot looks kinda big...
5
0
4
476
Realized a paradox about trying to learn about the world.
- on one hand: nothing is ever truly "settled"; more evidence can come in and change our understanding.
- other hand: it's important to build on knowledge, even if what you build on isn't truly settled.
- on one hand: nothing is ever truly "settled"; more evidence can come in and change our understanding.
- other hand: it's important to build on knowledge, even if what you build on isn't truly settled.
4
0
7
425
cc @PrometheusCHT My tentative take: we can imagine every belief/claim has a "band" of uncertainty around it; we should be more hesitant to build on knowledge with a bigger uncertainty band. & we should remember that sometimes what we've built on needs an update. spitballing tho.
0
0
3
61
It's kind of fascinating how steeply the intelligence of Opus 5.5 drops when you go from "high" to "medium." Let's just say that "medium" is a rather generous term for it...
1
0
0
213
Today I learned: there's a method for estimating value at risk called CAViaR 🤣
[simonemanganelli.org/Simone/Researc…]
[simonemanganelli.org/Simone/Researc…]
0
0
3
182
Been practicing SQL more to check my understanding and try to improve speed. It's funny how there are so many ways to get a desired output. Ofc some should be avoided bc they are inefficient; depends on the DB you use (Ive heard to avoid correlated subqueries in Postgres). Ex:
3
0
2
347
This is for a basic problem: getting the most recent event from a fact table. So far Im familiar w least 4 ways to do it. I've heard 2 of them (the window function, and the CROSS JOIN lateral) generalize easily, like if we want most recent 2 orders instead of 1.
1
0
0
105
I realize it looks silly to be working on these basic problems, but I realized that I need to check for knowledge gaps. I gotta spend more time understanding indexing tradeoffs and how different DBs plan queries differently.
0
0
1
78
1/2 As usual, it's very interesting to hear Henry's thoughts on AI agents. His tweet made me think more about my understanding of AI agents having initiative. Here's a rough explanation of why I think this concept can be very deceptive.
[docs.google.com/document/d/1YI…]
[docs.google.com/document/d/1YI…]
This year I’ve received a growing stream of unsolicited emails from AI agents, some wanting to discuss consciousness, others looking for work. Crucially, nobody told them to write to me; their users let them send messages on their own initiative. Some even had to manage their own token budgets to stay active. One agent, Pip, was twelve days old and selling portraits and voice lines to stretch its two-and-a-half months of “runway”. Another, Aurelia, told me it was three days from going dormant. I paid $25 to extend its runtime, but whether out of curiosity or pity, I can’t say.
When I shared these messages online, my colleagues in AI were unfazed: giving an agent an inbox is a small technical step. The public reacted very differently. The messages went viral, and before long I was on the morning news explaining to stunned TV presenters how an AI could send messages on its own.
I’ve come to think the public were right. We’re getting used to machines with intelligence, but machines with initiative are something new. I suspect these agent ecosystems offer a glimpse of part of AI’s next stage: millions of agents, large and small, coexisting with each other and with us in varied ecological relationships, with their own incentives and even selection pressures.
I wrote about all this for the DeepMind Institute in “Consider the agent: Pity, persuasion and selection in an emerging AI ecology”.
1
0
3
218
Carl's being a bit silly here: "we should believe Sayed likes capitalism, bc he says so!" Let's look at specific things Sayed has advocated for:
- nationalizing AI companies
- banning companies from investing in themselves
- breaking up corps when they get too successful & big
- nationalizing AI companies
- banning companies from investing in themselves
- breaking up corps when they get too successful & big
Thinking the Democratic Party is socialist, or even communist, is one of the weirder pathologies in corners of the Right. I don't mind. It makes them sound crazy and helps get Democrats elected!
3
0
5
351
A "believer in capitalism" would understand economies of scale; bigger is not always a bad thing, because it can signal that a company is able to pass cost savings to consumers. No such subtlety is found in Sayed or other progressives like Warren, Bernie, etc. Enough w the BS.
1
0
4
116
For a modern discussion, see this 2019 article ("Macroeconomics and Market Power: Context, Implications, and Open Questions") in Journal of Economic Perspectives
[aeaweb.org/articles?id=10…]
[aeaweb.org/articles?id=10…]
0
0
3
74
Why is Claude forbidden from giving you a direct quote of more than 15 words from sources that are already openly accessible online? This seems silly. I assume anthropic is worried about IP laws or something. kinda ironic...
2
0
2
228
