iiinova — where your digital self keeps evolving

Randy Boyes @randy.pub · 26/09/2026

100% - there’s a reason list-cols are a core tidyverse concept

Randy Boyes @randy.pub · 23/09/2026

My family had newer computers that I used before this but the first computer that was MINE was one of these Commodore XTs that I salvaged from my Dad's shelf of spare parts and busted computers

020

Randy Boyes @randy.pub · 20/09/2026

there are two (potential) effects on callback chance being investigated, resume quality and bias. In this example, the resume quality effect is high enough to overwhelm any effect of bias (there are two obviously qualified resumes and 6 obviously not), at least for the main result

081

Randy Boyes @randy.pub · 11/09/2026

I think there’s a different “speed utility function” for data science vs traditional comp sci. For a lot of data science applications, code that runs in under 10 seconds is basically as good as code that runs instantly - it’s when you start wanting to run that code 1000x that the problems show up

030

Randy Boyes @randy.pub · 11/09/2026

That being said the way R is typically taught does not emphasize writing fast code, and many people learn R as a first language and have no way of knowing what makes code fast

130

Randy Boyes @randy.pub · 11/09/2026

R is slow if you write slow R. Some of the simulation code I’ve reviewed is 10 or 100x slower than it needs to be, but if you’re copying your whole dataframe every iteration that’s just going to be slow in any language

120

Randy Boyes @randy.pub · 08/09/2026

No, it's something like: "Any data that can be used to distinguish or trace an individual's identity, either alone or in combination with other information." Lots of data that people generate online doesn't fit that, but yes, basically anything that describes a property of the individual in question

000

Randy Boyes @randy.pub · 08/09/2026

What? According to what definition? Both of your generic info examples would be PII in at least some cases.

100

Randy Boyes @randy.pub · 05/09/2026

Referring to bars as “hotels”

010

Randy Boyes @randy.pub · 21/08/2026

I’ve switched basically all my table to raw typst, either in a quarto typst block or in typst directly - most table creation packages in r and py are bad

100

Reposted by Randy Boyes

Presage Group @presage.group · 21/08/2026

New blog post: Reducing Human Error in Scientific Publishing Can LLM code review catch and reduce human error in scientific publishing? presagegroup.com/posts/science_erro… #analytics #ai #science

presagegroup.com

Reducing Human Error in Scientific Publishing

Can LLM code review catch and reduce human error in scientific publishing?

011

Randy Boyes @randy.pub · 19/08/2026

There have always been ways to complete homework “faster” - wolfram alpha, solutions manuals, get answers from peers, etc. I expect all of these, along with AI, are approximately equal in value for exam prep (zero)

110

Randy Boyes @randy.pub · 11/08/2026

don't believe his lies

000

Randy Boyes @randy.pub · 11/08/2026

Yeah at the uni I went to the process for doing this was literally called "the faint hope clause" and I don't know of a single case of it being used

170

Randy Boyes @randy.pub · 30/07/2026

This joke goes nuts if you, like me, are the son of a evangelical pastor who became an epidemiologist

3294

Randy Boyes @randy.pub · 23/07/2026

The gap between Design I and the rest is huge - its so much better it would be a shame if they chose a different one

130

Randy Boyes @randy.pub · 04/07/2026

I audibly boo’d when whatever game that was at SGF announced “Tupac” was going to be in it - I can’t imagine paying to attend an avatar concert

020

Randy Boyes @randy.pub · 01/07/2026

Ambitious but reckless junior engineer

110

Randy Boyes @randy.pub · 24/06/2026

If you want the answer to be more clearly “no”, maybe adding “in all cases” or “for all populations” or something similar would help

000

Randy Boyes @randy.pub · 24/06/2026

Isn’t this just a difference in interpretation of “overcome”? Obviously RCTs don’t let you observe both individual counterfactuals but they let you get an answer without having to - that could reasonably be called “overcoming” the problem

200

Randy Boyes @randy.pub · 20/06/2026

I’m an epidemiologist and I mostly agree with this - the majority of observational research you see doesn’t even have a question that’s asked clearly enough for the answer to be meaningful However - sometimes you can’t do the RCT, sometimes you can do good observational research, and -rarely- both

020

Randy Boyes @randy.pub · 13/06/2026

Anthropic’s marketing department must be ecstatic

071

Randy Boyes @randy.pub · 11/06/2026

Hate this (the linked doc) - Err on the side of giving credit, imo. Anyone who cares enough to stick with the project to publication and contributes ideas, text, work, and/or code should be in the author list

000

Randy Boyes @randy.pub · 09/06/2026

Everyone’s a descriptive until the implications section walks in

030

Randy Boyes @randy.pub · 04/06/2026

On paper, maybe its part of the job description... in practice, junior profs especially are paying a meaningful cost to review papers (their time) for little benefit - how many reviews would it take for that activity to be thought of as equivalent to one publication at a tenure review? =Inf, imo

100

Randy Boyes @randy.pub · 02/06/2026

For for figures, this should work already if you wrap and label them in a quarto figure block, like: :::{#fig-my-label} ‘’’{=typst} #image(“f1.pdf”) ‘’’ Figure caption ::: You can link/xref to the label with standard quarto syntax from there

110

Randy Boyes @randy.pub · 01/06/2026

I haven't used pdfpages but you can include external pdf files as images easily with `#image("a.pdf", page: X, alt="alttext")` if that's all you need

100

Randy Boyes @randy.pub · 29/05/2026

Wherever they are, I read them first, so I guess it doesn’t matter but it does feel odd

110

Randy Boyes @randy.pub · 28/05/2026

Actually spent multiple minutes this week wondering where the cal icon was

100

Randy Boyes @randy.pub · 27/05/2026

Sometimes you just need code that works and it doesn’t have to be gods gift to software engineering. If Linus is using it we can’t be claiming it’s only for people with low standards

020

Randy Boyes @randy.pub · 24/05/2026

I wish claude would just admit it doesn't know instead of attempting to gaslight me

023

Randy Boyes @randy.pub · 22/05/2026

You’re over the spec threshold where it’s worthwhile to run a local model for sure - I’ve had decent results on my 5070ti/64gb ram setup

010

Randy Boyes @randy.pub · 18/05/2026

BASIC in the 90s for me - not technically pre-internet but I was definitely learning out of a book!

010

Randy Boyes @randy.pub · 17/05/2026

These were some of the trickier statements to answer, but I chose to interpret “consensus” to mean “consensus based on the evidence”

000

Randy Boyes @randy.pub · 17/05/2026

Tough to answer these questions on an agree scale but the result is in the ballpark I think

010

Randy Boyes @randy.pub · 09/05/2026

Bruce only likes to chew toys while holding them over the edge of the couch, and gets very upset when he drops them

010

Randy Boyes @randy.pub · 05/05/2026

he adapted this from his morning affirmations didn't he

000

Randy Boyes @randy.pub · 03/05/2026

I know blue wins the polls online. That’s why I said blue wins when there’s nothing on the line The average number of lives saved depends on your assumptions about the actions of others, and I’m making different assumptions than you are

000

Randy Boyes @randy.pub · 02/05/2026

I agree it’s straightforward - blue will not get 51%, so some people will die. The only thing in your control is whether you die for no gain by picking blue or live by picking red

000