Hacker News: efskap

New comment by efskap in "LLM Voice"

efskap — Fri, 30 Oct 2026 00:18:46 +0000

How do you recognize sysco wrappers? Any particular signs to look out for? I guess the big one is a mismatch between the size of the menu and the size of the kitchen.

New comment by efskap in "Dots: Always-on agents"

efskap — Tue, 29 Sep 2026 18:20:26 +0000

But BERT's name led to some wonderfully named spinoffs like CamemBERT and FlauBERT for French, as well as ALBERT, RoBERTa, etc.

New comment by efskap in "Google’s Project Suncatcher to put ML infrastructure in space"

efskap — Fri, 25 Sep 2026 00:46:46 +0000

Right, the goal here is a more realistic Dyson swarm, not a Dyson sphere (yet)

New comment by efskap in "Claude discovers a novel enzyme system with CRISPR-like repeats"

efskap — Wed, 23 Sep 2026 23:01:05 +0000

Predicting reality, under the "controlled hallucination" framing, corrected by sensory error signals. The brain has no access to ground truth, only input data that helps correct the hallucination.

New comment by efskap in "Claude Code from Source"

efskap — Fri, 18 Sep 2026 01:58:01 +0000

Easier than reading the source myself, but I don't think the site makes a good case for why I SHOULD apply CC's patterns to my own systems, besides it being the most used harness. It's clearly useful software, but that doesn't _necessarily_ translate to something to learn from.

It is apparently a vibecoded mess: https://news.ycombinator.com/item?id=47772282

New comment by efskap in "Tiny $70 Xteink X3 e-reader"

efskap — Mon, 14 Sep 2026 21:54:32 +0000

It's an additional layer on top of the eink that has to diffuse light from the side. It's bound to reduce contrast at least a little

New comment by efskap in "NEO Emacs – GPU-Accelerated Emacs Powered by Rust"

efskap — Mon, 14 Sep 2026 21:02:26 +0000

There's an interesting prompting technique for this that asks the model to start with a random string that is then manipulated into the answer: https://pub.sakana.ai/ssot/

New comment by efskap in "Ask HN: Why were OpenAI, Claude, and Grok simultaneously down?"

efskap — Thu, 03 Sep 2026 18:31:32 +0000

This is like the Bronze Age collapse when city-states fell one by one to displaced demand, under the refugee interpretation of the Sea Peoples.

New comment by efskap in "Claude Fable 5.1 and Claude Mythos 5.1"

efskap — Thu, 03 Sep 2026 00:19:06 +0000

Only in South Asian varieties of English apparently: https://en.wiktionary.org/wiki/of#Pronunciation

New comment by efskap in "Claude Fable 5.1 and Claude Mythos 5.1"

efskap — Wed, 02 Sep 2026 01:26:41 +0000

They sound exactly the same to me.

Wiktionary gives as /ˈʃʊdəv/, unstressed as /(h)əv/ and unstressed as /əv/.

New comment by efskap in "“I just chose words carefully”"

efskap — Tue, 01 Sep 2026 06:33:37 +0000

The best of both worlds would be achieved by elastic tabstops. https://nick-gravgaard.com/elastic-tabstops/

If only they took off.

New comment by efskap in "Google Antigravity introduces Boost deep reasoning (/boost)"

efskap — Tue, 01 Sep 2026 05:55:17 +0000

And these are the guys who invented transformers! How?

New comment by efskap in "“I just chose words carefully”"

efskap — Mon, 31 Aug 2026 06:27:38 +0000

Well, estimating string length of tokens is much like counting R's in strawberry.

New comment by efskap in "M5Stack Launches PaperMono"

efskap — Fri, 28 Aug 2026 01:36:39 +0000

There's a reason it stands for Bring Over Another Thousand

New comment by efskap in "Show HN: I made a Raspberry with Qwen my local car AI"

efskap — Wed, 26 Aug 2026 00:20:34 +0000

Yeah referencing is the way to go, as even finetuning probably captures style more than concrete facts. I know with large context windows we don't really RAG anymore, but for owner's manual lookup with a smaller model it seems ideal.

Something every LLM user ends up learning is that they're far better used as search and summarization tools than as knowledge databases in themselves.

New comment by efskap in "Claudette: Make Claude stop talking like a BuzzFeed article"

efskap — Sat, 22 Aug 2026 01:41:26 +0000

But comments should hopefully explain "why", not what or how, and that kind of context is exactly what an LLM is likely to be missing

New comment by efskap in "How Bluesky draws its logo on screenshots"

efskap — Tue, 18 Aug 2026 00:16:40 +0000

I guess it would need permissions, but you can monitor the photo library with PHPhotoLibraryChangeObserver and check if a new item's subtype is photoScreenshot.

Otherwise userDidTakeScreenshotNotification only fires for your own app

New comment by efskap in "How Bluesky draws its logo on screenshots"

efskap — Tue, 18 Aug 2026 00:09:32 +0000

The app actually is notified after the screenshot is taken: https://developer.apple.com/documentation/uikit/uiapplicatio...

Android does it too: https://developer.android.com/about/versions/14/features/scr...

New comment by efskap in "The Indo-European Family Tree"

efskap — Thu, 13 Aug 2026 21:59:26 +0000

Yeah, it's not a clean inheritance but lots of linguistic contact and negotiation.

And crucially it's not just other IE-adjacent languages but the Pre-IE substrate. So many Greek words are inherited from the mysterious languages that were there before, of which we only know through the marks they left on ours.

https://en.wiktionary.org/wiki/Category:English_terms_derive...

New comment by efskap in "Where did the old web go? We followed 657,607 links to find out"

efskap — Thu, 13 Aug 2026 18:37:23 +0000

It's the great irony of digital media. Copying data is accurate to the bit and is preserved "as-is", but in practice, it requires someone to maintain servers, to care about it. To separate out what is worth preserving.

We hot-linked to all those image hosts because we couldn't imagine them disappearing.

Archive.org had incredible foresight and if it didn't already exist, I'd call such a project a pipe dream.