Mastodon Feed: Posts

andrewnez ("Andrew Nesbitt") wrote:

New hobby: trying to make deterministic versions of llm powered dependency analysis tools using the various parts of git-pkgs.

Normally can get 90% of the way that and it ends up being much faster and cheaper.

Boosted by baldur@toot.cafe ("Baldur Bjarnason"):
jk ("josef") wrote:

something the whole "AI" thing has taught me, is that Time is the main ingredient to making things that are good. anything which 'boosts velocity' or 'makes you 10x more productive' reduces the amount of time you use while making it. it's not that you're 'saving' time, you're just refusing to use as much of it now for some reason. you recognize it's a valuable ingredient, so you want to conserve it: to keep it in the tin rather than actually add it to the recipe. and then the result tastes worse

Boosted by baldur@toot.cafe ("Baldur Bjarnason"):
lea@ordinary.cafe wrote:

If you/your project is affected by DigitalOcean stopping their OSS credits program, please shoot me a PM immediately. Might be able to help.

Boosts for visibility would be appreciated.

#foss #linux

ireneista@irenes.space ("Irenes (many)") wrote:

it doesn't feel that long, but also we for sure cannot complain, it is a year's worth of functionality

ireneista@irenes.space ("Irenes (many)") wrote:

it just occurred to us to wonder how long we've been working on Evocation

the git history says it's exactly a year. like today is the one-year anniversary. neat!!!

dysfun@treehouse.systems ("gaytabase") wrote:

there is something special about idris-mode that makes it pop up messages in the absolute least convenient place. like the buffer you were just typing into.

brennan@social.lol ("Brennan Kenneth Brown 👻 BOO!") wrote:

Here Is a Person: Enter the Tildeverse

https://brennan.day/here-is-a-person-enter-the-tildeverse/

#smallweb #history #linux #unix #community #technical

baldur@toot.cafe ("Baldur Bjarnason") wrote:

“Dining with wolves – Notes from a data witch”

https://blog.djnavarro.net/posts/2026-10-05%5Fdining-with-wolves/

> Ideology is dark, it is violent, and it is not built atop reasoned beliefs. You cannot have a polite conversation at the pub with someone who openly wants to kill you.

Boosted by baldur@toot.cafe ("Baldur Bjarnason"):
thomasfuchs@hachyderm.io ("Thomas Fuchs 🫯") wrote:

The best single book about all of this remains “Computer power and human reason: From judgment to calculation.” by Joseph Weizenbaum.

It’s 50 years old this year, and obviously doesn’t cover LLMs specifically, but the concepts apply.

(You can easily find a PDF of it online.)

Boosted by baldur@toot.cafe ("Baldur Bjarnason"):
timnitGebru@dair-community.social ("Timnit Gebru (she/her).") wrote:

For the uninitiated, Bernie Sanders' AI bills are being written by a eugenicist cult full of predators organizing "rape orgies."

Michael Vassar in that list is also the former president of MIRI, the organization closely advising Sanders.

Guys, we can't except our "leftist" politicians preoccupied with the 1% and all that to care about eugenics or sexual assault or cults or anything like that right?

Boosted by baldur@toot.cafe ("Baldur Bjarnason"):
mcc wrote:

Categorizing slop by how wet it is

Boosted by baldur@toot.cafe ("Baldur Bjarnason"):
patigallardo.bsky.social@bsky.brid.gy ("Patricia Aas 🐢🏳️‍🌈") wrote:

I’m just sayn’, don’t say I didn’t warn y’all.

It’s not my fault people just can’t bring themselves to believe that a whole slew of Silicon Valley Tech Bros are self centered nutjobs.

RE: https://bsky.app/profile/did:plc:ybtyn5l4nljys46ijqtpldaw/post/3mkb27axxts2e

Boosted by baldur@toot.cafe ("Baldur Bjarnason"):
patigallardo.bsky.social@bsky.brid.gy ("Patricia Aas 🐢🏳️‍🌈") wrote:

My most evergreen post.

baldur@toot.cafe ("Baldur Bjarnason") wrote:

RE: https://fedia.social/notes/aryl69qiee7spmvn

The problem with Ben Thompson saying this is that his entire body of work, his perspective, advice, and theory of the corporate, is geared towards this result. Very few people have done more to outline the industry’s orthodoxy in coherent language and it’s exactly that outlook that’s creating the outcome —letting the incoherent and oblivious march forward under a single banner in the unified direction of crap and broken buttons

ireneista@irenes.space ("Irenes (many)") wrote:

anyway, yeah. that was super easy, worked first try even. we'd figured it would be, it just didn't make sense to do it while we were working on things that would be directly affected by it...

Boosted by baldur@toot.cafe ("Baldur Bjarnason"):
iris_meredith ("Iris") wrote:

I've written a new article: this one started off when I read @pluralistic's excellent book Enshittification, and it's my attempt at an analysis of what enshittification looks like from the inside out. Why do people enshittify? What set of beliefs and mentalities make people able and willing to enshittify? And why has the outcry over it been comparatively muted?

https://deadsimpletech.com/blog/poor-enshittified-souls

ireneista@irenes.space ("Irenes (many)") wrote:

our whole hex-transform strategy means that you can, in fact, check out exactly how it's implemented and audit it. the ultimate root of knowledge is the hex dump, not the binary.

this was of course the whole point of our strategy, it's just a particularly nice example of how pleasant it is in practice

ireneista@irenes.space ("Irenes (many)") wrote:

this is the thing that Reflections on Trusting Trust references, where if you add a new literal syntax and then start relying on it, you wind up with a compiler that can do the new thing, but in a sense the knowledge lives only in the binary... the source can't be compiled without already having a binary that can do that

except

ireneista@irenes.space ("Irenes (many)") wrote:

and here's the fun part

we added the syntax in https://code.irenes.space/evocation/commit/?id=589e56e26b8e997fa748ccc8fcd5b4bfcd2241a7

then we immediately made use of it, rewriting all the existing references, in https://code.irenes.space/evocation/commit/?id=ec7fefbae7b9651b5cfe320b2d8b8db27fbc683f

ireneista@irenes.space ("Irenes (many)") wrote:

so we went ahead and create the syntax x' which does that in one step

"x" for "execution token"

ireneista@irenes.space ("Irenes (many)") wrote:

we'd been noticing a pattern, ever since we first started writing the transforms, where sometimes we want to do that and then immediately go from the entry pointer to the execution token

which, because we're not fond of most abbreviations, was quite verbose

dysfun@treehouse.systems ("gaytabase") wrote:

littlegem: so we have co-debruijn well-scoped syntax now. getting there from the presyntax was a bit more code than i'd like, but it's all fairly paint by numbers once you've worked out the basic approach.

co-debruijn is a little different from debruijn - terms are indexed by the names they actually depend on rather than what's in scope, so it means we've whittled down everything to just what matters. as part of the process, we instantly notice things like functions that ignore their argument because it's required to make the types check. and we never need to ask for a list of used variables, because terms are naturally indexed by them. kinda neat!

this seems like a much better place to start nbe from

ireneista@irenes.space ("Irenes (many)") wrote:

sweeeeeet

so you know how ' is a special word that lexes the name of a word and acts like a literal reference to that word's entry pointer?

slightlyoff@toot.cafe ("Alex Russell") wrote:

And I don't mean "harder" in the sense of putting the semicolons and braces in different locations; refactoring has never been more automated.

Thing have gotten harder to fix because entire layers of management are now fundamentally unfamiliar (best case: rusty) with the evidence-based decision-making necessary to nurture a healthy product culture. Some never got promoted for doing an objectively good job; starting from zero. It's pushing molasses up a hill with this sort.

slightlyoff@toot.cafe ("Alex Russell") wrote:

RE: https://social.vivaldi.net/@brucelawson/117387522243320669

While I can't claim any credit for the GH move -- their choices appear to be a random walk through a museum of footguns -- I did give an internal talk earlier in the year titled "CSS-in-JS: Frontend's Worst Idea?". You can probably predict the contents: massive memory and processing overhead, style invalidation a-go-go, etc.

What's worst about this entire debacle is the gutting of engineering culture. Bros picking stacks on GH stars and twitter whuffie. Every repair is harder now.

ireneista@irenes.space ("Irenes (many)") wrote:

we need to teach it about different kinds of executable image at some point soonish - there are things people might wish to compile that aren't the compiler.

we kind of feel like doing some easier features first though, there's a backlog of those...

ireneista@irenes.space ("Irenes (many)") wrote:

there we go. we just checked in a working version of the enchant/inscribe thing that we started on... two weeks ago? before we caught COVID, anyhow

db@social.lol ("David Bushell 🪿") wrote:

"The Wheel of Time turns, and Ages come and pass, web developers are surveyed, leaving memories that become legend. Legend fades to myth, and even myth is long forgotten when the Age that gave birth to it comes again."

- Robert Jordan

Boosted by andrewnez ("Andrew Nesbitt"):
civodul@toot.aquilenet.fr ("Ludovic Courtès") wrote:

#Guix gained the ability to make your channel file essentially an alias for a remote channel file:
https://guix.gnu.org/manual/devel/en/html%5Fnode/Downloading-a-List-of-Channels.html

One use case is to delegate the choice of channel commits to pull to a continuous integration server.

Another one is, within a development team, to delegate that choice to one person within the team, so that everyone downloads a “known-good” set of channels.

db@social.lol ("David Bushell 🪿") wrote:

apparently this was unknowingly merged and before an explicit "No-AI" policy was introduced 🫤 oh well