Simon @spoltier@qoto.org

So… retired/unemployed programmers or employed ones who’d like to contribute to open source:

What do you think of a “Civilian Documentation Corps”¹?

There are tons of contributed packages to popular languages that are fine but underused because underdocumented. Some of the package owners are well-intentioned but no good at documentation or don’t understand its importance. Why not swoop in with an offer of help? (1/6)

rage against the state machine

I love raspberries, but artificial raspberry flavoring just tastes like death 🥲

@kaoudis I always thought the artificial banana flavoring was weird, until I tried a ripe (black /brown) banana...

#notesArt #mastoArt #art #illustration

Scott Aaronson's blog is always good reading; in this case, it's linking to other good reading. Excerpt:

"We’re still treating this as a business and technology story like personal computing or the Internet, rather than (also) a national security story like the birth of nuclear weapons, except more so. And we’re still indexing on LLMs’ current capabilities (“fine, so they can pass physics exams, but they still can’t do original physics research“), rather than looking at the difference between now and five years ago, and then trying our best to project forward an additional five years.”

https://scottaaronson.blog/?p=8047

This talk by @maggie was a real highlight of the local-first conference. A compelling argument, very well articulated https://maggieappleton.com/home-cooked-software

#llm I like saying weekend project, although tbt, I think the past couple years have been a series of "weekend projects." Anyway this weekend's project was benchmarking LoRA fine tunes with torchtune, axolotl, and unsloth w/ 4090, 3090, W7900, and 7900XTX cards. Besides getting some practice setting up configs for the different trainers, it was good to poke around w/ wandb a bit more for report publishing): https://wandb.ai/augmxnt/train-bench/reports/torchtune-vs-axolotl-vs-unsloth-Trainer-Comparison--Vmlldzo4MzU3NTAx

I am once again reminded that in the fediverse you never know what somebody is really responding to because you never know what's visible or invisible to you. Anywhere.

#federation #federationfail

A discussion of discussions on AI Bias:

https://danluu.com/ai-bias/

So it really is a time to be curious and not contemptuous in my opinion. There is a tendency to say either that tech is sacrosanct and cannot be criticized or tech is evil and must be obliterated. I pretty much felt the same working with the public education system. Tech is people, at least the parts of it that I am interested in engaging with and for whom I think applied psychology can be a benefit. Some compassionate curiosity about WHY we see utility in "contest cultures" can be very helpful

I'm very into learning programming languages:

a) kind of “badly” — often never learning major features or major parts of the ecosystem, or not using very popular tools
b) with a lot of confidence -- where I feel 100% confident in the limited subset of the language that I do use
c) over a long time -- sometimes only starting to use a 'basic' feature maybe 5 or 10 years in

4/?

Jointly pre-defining what qualifying is, is a known protection against bias later in the process. I think that it's an underappreciated part of this whole thorny problem, the fact that the END of a hiring process can start to warp and introduce uneasy and invisible bias as people start to parse things as differences under the pressure of decision making. Same with grad school admissions imo: it's foolish to act like we are identifying the ten best when the top 100 are completely comparable.

It's also just a way that some areas rigor-wash their terrible, terrible practices that are failing the majority of students. It's maddening to see the status games that some fields play to assert they are too smart, technical, and unique to do things like "be able to teach all the students this university serves not just the ones hand selected by you to be just like you"

As a postscript I've been working on tying together the experience of advocating for change with learning science in public schools and nonprofits and my work on ability beliefs on software teams into a talk I have called "Fighting Dirty for Good Culture" and it has not yet found a home -- would really love to give it at a tech conference sometime 🙌

@b0rk is Python 'worse'? not in any way I can't work around with some elbow grease (for example every library entry point has dynamic handwritten type checks as a policy)

is Python 'too slow'? i recently implemented a naive userspace networking driver in Amaranth and asyncio. I could hit 80 Mbps and some tens of kpps with three days of effort and very little optimization

choice of language is a tradeoff and most people present it as a far more dramatic one than it actually is

#notesArt #mastoArt #art #illustration #typography