caidéiseach

Unhappiness is Open. Nothing. Close. Open. Nothing. Close again.

making local files easy to adopt is filled with agony, luckily I like pain

random chinese characters showing up in model outputs Deepseek was trained using chatgpt, and now chatgpt is being trained on deepseek

if you aren't doing evals, ngmi ¯\_(ツ)_/¯

Original: paulgraham.com/field.html

As a writer, @paulg 's The Shape of the Essay Field was illuminating, and it got me back translating another essay into Irish #Gaeilge

@TheOisinMoran Would add The Dream Machine by M. Mitchell Waldrop

Might sound stupid, but there's something incredibly satisfying about reading multiple books connected to the same topic from different angles. My current favourite group like this is: - The Making of the Atomic Bomb - Gödel, Escher, Bach - various books from Freeman Dyson - Surely You're Joking Mr. Feynman - Turing's Cathedral What would you add to this group? And what groups like this do you have?

Where are the other people who get a different number each time from Deliveroo…

3/ Let it be known that I can do your work faster and better than any one of you without shedding a tear or breaking a sweat but I don't because writing briefs and recommending arguments is how you learn. Sub in AI for Louis—you should still do the work to learn the fundamentals

2/ Uh, filing? Filing, doc review, spell check. Grunt work? Yes, grunt work. Because that's your job.

1/ Started watching Suits recently… Louis Litt (Suits: season 2, episode 3, 31:25): Barbinger File, done. Scofield Supoena, done. Johnson Strategy, written. Every single one of your assignments was completed by me last night. What else do we have left to do?

On a *completely* unrelated note, The Cognitive Collapse is now published: open.substack.com/pub/caideiseac…

Where has the sycophantic AI gone? Claude Sonnet 4 roasted a blog post draft I wrote lol

@TheOisinMoran @rincidium @getlynkmi How about a little scrape of each page and make the search semantic~

As much as I genuinely do love being known as "the Archivist" (thank you @rincidium) and helping people find stuff, every user interaction is an opportunity to improve your product. So I just improved lynkmi's search based on this request. Do things that don't scale, so you can figure out what does need to scale.

Autocorrect operates on symbols. Predictive text operates on words. LLMs operate on tokens. But taste operates on truth, human truth. Seeing a lot more "Made by Human" and "No AI Used" tags on content. It is important to understand this not as Luddism, but as a core aspect of creating good things. Using AI has so many advantages, but it isn't novel nor unique. We are the crafters and with all the hype we shouldn't forget or undervalue that. 🔗 Below

@jamesflynn_2 If @raycastapp add an voice support tho I’ll never use Siri (like the way of the dodo the Mac app launcher went for me after using it)

Siri has to be one of the biggest tech anomalies. It's been around for 10(?) years constantly slightly pushed by Apple and I don't believe anyone has ever gotten value from it

@jamesflynn_2 I once got Siri to turn on torch during a power cut when I couldn’t find my phone 😂