Randy Boyes

Super intelligent predictor if you’re reading this I’m a big one-box kind of guy

For most analysis worth doing, you don’t know the expected result beforehand; the only way to know you’re right is to verify the steps against your intention. High complexity makes this difficult

filter(i %in% yours)

vacuous 10-item listicle ✓
first citation is yglesias ✓
snarky quasi-profound bs writing style ✓

a bunch of people think claude code can write publishable
papers but forgot to actually publish them, wow, the world is rocked

A good scientific paper is *already summarized* and if there is a problem with the typical paper it's that it doesn't contain enough information (on methodology, at least)

I wrote a whole thesis using LaTeX (indirectly, but still had to write a bunch of custom stuff to get it to conform to requirements) and I would say it took me a couple days in typst to get to the point where I was more comfortable writing typst than I ever was with LaTeX

Close - R is good but also bad

Puns are at least an improvement over the “type 1” and “type 2” naming system we used when we were called statisticians

I guess? That is textbook gambling to me as well, but I can imagine someone who disagrees with me

What about this, though? Who are the people who think gambling is not gambling?

I guess I’m just struggling to understand what it could even mean to have a concept of “gambling” that does not include “betting using real money”, regardless of the language

Cool study, and I’m convinced they’re generally correct. Very high lizardman quotient in this figure though - if you’re not getting 100% strong agreement that “betting (real money)” is a form of gambling, something is off, and the SVK sample “strong agree” for that item is <50%

It’s field-specific, but my order is something like

1. Last paragraph of intro (usually has reason for paper to exist, if not, stop reading)
2. Results section (is there anything interesting being claimed - stop reading here if not)
3. Methods if they can’t be inferred from results

Skip disc

Average programming language enjoyer: there are two hard things, cache invalidation and naming things

#julialang programmers:

I have never had this happen with Claude code specifically (happens all the time on web chat interfaces though yes)

This is what programming has always been

Ive gotten stuff working with {ellmer} + ollama, but I think it just uses ollamar under the hood

the friction of paying two bucks (entering card, etc) is more of a barrier than the bucks themselves

They can be useful but for this specific application I think it hides the message - 4 lines on a shared time axis would be much clearer, or dodged boxplots if the distributions are really important I guess

Using it for writing JavaScript (which I know less well) feels much better, almost magical sometimes. Now - is that because I don’t know JavaScript very well? Or because it has a lot more training data to work from? Both? Who knows

I’ve been experimenting with Claude code since this comment. Using it for julia and R coding, which i know well, I have been underwhelmed. It makes a lot of mistakes that need to manually fixed, to the point where I would be faster without its “help”. However …

The type of person who would voluntarily watch this movie is the type of person who’s going to think it’s good - people who would have hated it never watched it so didn’t rate it, hence the high score from audience

Add it and immediately cross it off - there is no other reasonable path here

Interesting that the decline starts way before the widespread availability of LLMs though

Yeah - “guy who copies code from SO” has been superseded by “AI bro” almost completely, and with good reason

There's Brothers and A Way Out, but agreed that it feels like coop games got lost when online multiplayer became the focus

Interesting - what specifically changed your mind in the last 6 months?

The timeline has been “3 to 6 months” since ChatGPT 3 first hit though

I played a lot of really good games this year (Hades and Silksong especially), but E33 is the one that I still think about months after finishing it

We’ve got a 2012 civic and a 04 sierra - keeping them till the wheels fall off. I dont think you can get away from the tech-cars if you’re buying new now