🇺🇸 “In 1842, a young journalist interviewed 91-year-old Levi Preston, the last survivor of the battles of Lexington and Concord [and] expected to hear stories about unconstitutional government and oppressive taxes.”
Remote
Dave
@daveliepmann@social.tchncs.de
Lastenrad dad techbro in Berlin.
174
Followers
199
Following
31
Posts
Joined April 03, 2017
Posts
Currently enjoying the "text extruding" LLM metaphor.
Serve words like soft ice cream from a tap
Dairy sourced from the Great Milk Sea, formed by pumping the teat of ~all human writing
Ancient taboo prohibits going near the milky reservoir ourselves
Instead we fitted every nook and crevice of it with hoses, each leading to a sort of N-inputs Chinese room
Flavor is steered by a leprechaun in the room, doing his best to pick the right combination of hoses according to your instructions
but sometimes he loses his reading glasses
It's malpractice to write about giving up running because of injuries and not even mention strength training.
It can't solve *everyone's* problems but it can solve a lot, and we need to crush the childish belief that just doing the sport itself (especially one as brutal as marathon+ distance running) is good training.
Scalable OLTP in the Cloud: What’s the BIG DEAL?
Ceci n’est pas about LLMs
1. Develop a broad, varied base of knowledge
2. Load context by closely reading a small corpus, developing models of what they say and how they relate to each other
3. Clearly & narrowly state the problem
4. Set constraints on the solution
5. Decide how you will evaluate the solution
6. Explore solutions in parallel, being careful not to let the assumptions from one solution-exploration poison the others, but also surfacing discoveries between them
7. Check your work rigorously
Some nice words about LLMs pushing us apart
https://notes.piercegleeson.com/researching-on-a-post-reality-internet/
Replying to
@daveliepmann@social.tchncs.de
Replay is pretty sick too. Refusing to acknowledge head position or the overhook belt grip paid off.
His secret? Just be stronger and have better footwork.
@gregburd@mastodon.social I’m getting 404 on https://github.com/gburd/pg_mentat — is it perhaps still private? The announcement is intriguing :)
Replying to
@marianoguerra@hachyderm.io
@marianoguerra@hachyderm.io “Components are the subject of composition” is an elegant framing. And as a bonus it leads to quiche-is-a-sandwich mind-bending edge cases!
"Somewhere through this web of satellite trails is Comet C/2025 R3 (PanSTARRS)..."
LLMs enabling a developer to switch from emacs to vim might be the technology's greatest crime
Replying to
@heals@indiepocalypse.social
@heals@indiepocalypse.social @murena@mastodon.social @helma@mastodon.social sounds like an opportunity for @sovtechfund@mastodon.social
Chouser released a beautiful UI over the very beginning of Clojure on IRC
Replying to
@plexus@toot.cat
@plexus@toot.cat I choose to believe that writing for humans will continue to be valued. Keeping the torch lit.
Replying to
@plexus@toot.cat
@plexus @jack One recent project is instructive w/r/t productivity & task suitability.
I forked a small Clojure thing someone else wrote. It's large enough that it would take me at least a half-day to read before making changes. It does X to Y and I want X for Z. Estimates are hard but coding that port manually would be, I dunno, at least a few days of work.
With an LLM I *think* of a natural-language specification, *plan* how I would accomplish the task, outline how to accomplish those goals (as if to a peer or a slightly junior colleague) and Claude executed it in minutes. I can dip into impl but don't have to.
Then I intentionally used the LLM to "declaratively" tweak organization/presentation of content, which 100% would've been faster & better if I'd done it manually in emacs.
Replying to
@daveliepmann@social.tchncs.de
@plexus @jack
Satisfaction? Pride? I'm grieving the sudden irrelevance of my craft. At the same time I'm excited to learn this "ebike for the mind". It lowers the effort to produce things I want. It opens doors to software projects I don't have time for. Craft is good but not every woodworking project needs to be hand-whittled. (And both skills can/should coexist!)
It's satisfying to build things. Pride shifts somewhat to the product side, the taste/experience/architectural chops to know what to have the LLM write.
Replying to
@daveliepmann@social.tchncs.de
@plexus @jack Do I understand the code? Depends. Sometimes I could but choose not to because it's not important to the task, which feels like declarative programming. Writing elisp with LLMs is often like that for me. When the impl matters then yes I take the time to review it as if I or a colleague was writing it in the before-times.
Replying to
@plexus@toot.cat
@plexus @jack "AI mandates are pretty common, so clearly some people need to be coerced." To me this is like "the popularity of ALGOL languages reflects their superiority over lisp and ML language families". Social dynamics are their own weird phenomena.
This is not the only instance of LLM skepticism rising to the level of interrogating whether I'm being honest with myself. I find it rather rude.
I've used it for a variety of projects, I've seen my colleagues do incredible things with it (including one-shot debug) on extremely complex projects, it's changed my relationship with my emacs config. I think LLM workflows are still deeply in flux and actively being discovered, and there's worth in several kinds (interview, "pairing", agentic, multi-agent) depending on the task.
Replying to
@plexus@toot.cat
@plexus @jack as someone who is simultaneously mourning the lost craft and excited to learn a powerful new tool, I don't think the token quotas are the problem. The problem is the tool works. The marketing hype is still BS of course but LLMs really do mechanize the production of a great deal of code.
Replying to
@daveliepmann@social.tchncs.de
@jstepien@mastodon.social ESPECIALLY that the URL eats it with no problem