Miles McBain

The human skin toned squiggly asterisk logo for Claude.

Hey Claude

Draw yourself a logo befitting The Suppository of All Wisdom.

This man spending just 0.5K a month on tokens and working only on his nights and weekends has shipped an application every 30 days.*

*Individual results may vary

When we talk about Baby Boomers not getting off the stage: In 1997, the US president was born in 1946.

In 2007, the US president was born in 1946.

In 2017, the US president was born in 1946.

And next year in 2027? The US president will have been born in 1946.

The analogy is 💯

It’s all the pain of raising the small child with none of the reward. Lil Claude only learns as far as his context window allows. After that he resets to the mean opinion of all the text on the internet. 😱

Parenting with an extra Sisyphean twist.

a woman in a bathrobe is standing in front of a painting Alt: Fran Fine in a bathrobe is standing in front of a painting and opening her arms wide as the camera uncomfortably zooms in to her face

Sorry, Bluesky, but I have to say it: The little gremlin in my brain would not stop screaming about AI until I referenced Fran Fine and linked to "Me and You and Everyone We Know." erincikanek.com/the-rumors-o...

👏👏👏 Yes!

folks, please don't submit LLM-generated PRs to open source projects. It makes no sense.

If the maintainers want to use an LLM to fix an issue, they can use Claude or whatnot directly. They don't need you as intermediary, that's just silly.

If they don't want to use LLMs, they have reasons.

But at least with ggplot2 the ‘+’ operator establishes something unusual / magical is happening. Pipe interface strongly implies data flow to me.

Yeah. I’ve had some success with it and would use it again, but I deeply feel the cringe in your workflow() -> fit() example. You’re not executing functions on data, you’re building a specification to do that at some later time in an unspecified way. Similar to ggplot2 in that regard.

Feeling safe as a statistician because no AI can respond to a question like we can. Show me an AI that can gently rub its temples and say “it depends” before taking your ass on a journey of technical details.

Data.table’s weakness is being overly focused on a particular end user experience. The rough edges show up when you want to use it at as a platform to develop your own user experience.

Yo where’s the targets zealot phase?

That’s unexpected. In VSCode the terminal working directory is inferred from the active editor. I’ve never wished it was any other way.

Haven’t used positron but I do this in VSCode a bit. Quite nice if you have internal packages supporting projects.

I wrote up something about it ages ago milesmcbain.micro.blog/2022/10/18/a...

Are there any R users working for the Australian federal dept of health? I’d love to hear from you!

New gig, new people to meet :)

#rstats 4.5.3 "Reassured Reassurer" scheduled for March 11. Full schedule on developer.r-project.org (or the svn if you're impatient.) This should be the wrap-up release for the 4.5 series.

So, what's the next #rstats flame war on the schedule? Base |> vs. magrittr %>%?

Of course if you shift the goal posts enough, you can argue we already have both.

Yet the truck drivers remain as employed as software engineers.

I’ve started referring to the idea of LLM agents that can magically build complete working software applications from vague human specifications as “Full-self-coding”.

I think it’s on the same time horizon as Full-self-driving. I.e. likely never.

GitHub - MilesMcBain/breakerofchains: Break your chain at the cursor line. Run the first bit. See the output. Be free. Break your chain at the cursor line. Run the first bit. See the output. Be free. - MilesMcBain/breakerofchains

Introducing probably my most used keyboard shortcut

github.com/MilesMcBain/...

every now and then I think folks in #rstats need to be reminded of or introduced to browser() and debug()

this rage tweet brought to you by The Millennials™️, many of whom physically wired up their teenage homes, were introduced to IP addresses with 192.168.1.1, and have this particular shade of blue burned into hippocampus

Python is slipping in popularity – Tiobe The world’s most popular programming language is losing market share to more specialized languages such as R and Perl, Tiobe says.

#RStats is up baby

www.infoworld.com/article/4129...

It’s definitely worth it to do some base-only projects. Packages make good candidates due to discrete nature, limited scope etc.

Since things tend to blow up, you can pin them down pretty quickly with debugging tools.

On the plus side these failures can often highlight areas of fragile code, as they occur in situations that are susceptible to common edge cases.

It's an unfortunate situation for sure.

However, due to the nature of the changes, things will typically totally fail, rather than introduce some subtle issue that leads to an incorrect result. And I am sure this was considered in the design of tibble.

No denying that.

Though *in practice* I find they behave more in line with my expectations for this type of object.

I don’t have to remember the edge cases that feel quirky to me. Quirks that make sense in a different context, I am sure.

*largely.

The sweat and tears I have shed in the service of data frames!

Dirk at this point I’ve been programming for nearly 30 years. 10 in R. The idea that my opinion is based lately in theory without consideration for practical issues. Wow.

🤭🤭🤭