Hacker News: nomel

New comment by nomel in "How AI text watermarking works"

nomel — Fri, 14 Aug 2026 01:17:31 +0000

The tech bro speak gets so insane that, if I paste it to another claude session, it's so unintelligible that new claude session doesn't even know what the other was trying to say.

It'll be interesting to see if all this causes a subtle shift in vernacular.

New comment by nomel in "Choose Boring Technology (2015)"

nomel — Thu, 13 Aug 2026 23:41:13 +0000

Sounds familiar! I believe it was a simple html page hosted on some university user page.

I've tried over the years, but can never find it. All the search engines moving to vector search makes finding something like this near impossible.

New comment by nomel in "A small battery has been ringing a bell since 1840. No one knows what's inside"

nomel — Thu, 13 Aug 2026 20:35:26 +0000

It would also be funny if someone was secretly in charge of maintaining a magnetic coil built into the shelf under the based, as a very long running prank.

New comment by nomel in "Choose Boring Technology (2015)"

nomel — Thu, 13 Aug 2026 18:10:13 +0000

There was another, much older, post similar to this about, I think, "beans" that engineers use to solve problems. If I remember correctly, it was something like, solving a problem costs "beans", and engineers will always use most all of their "beans" to solve a given problem, because it's somewhat "easy" until you run out of them.

Maybe it wasn't beans? But, I've been looking for it for years.

New comment by nomel in "Pixel Watch 5"

nomel — Thu, 13 Aug 2026 16:48:44 +0000

Here we go: https://feng-qian.github.io/paper/watchenergy_mobicom16.pdf

If you turn off cellular and WiFi (making it pebble equiv, but not gaining you weeks of battery), it's mostly display and CPU. I'm guessing a good portion of that CPU usage is display related. Pebble uses e-ink.

Maybe there's really that much more going on with Android?

New comment by nomel in "FDA Admitted in Print That Nothing It Knows of Will Wash Cyclospora Off Lettuce"

nomel — Thu, 13 Aug 2026 00:30:29 +0000

Where I am, all the farmers markets are full of produce marked with the local grocery chain, or boxes from mexican grocery chains. I suspect they all "fell off the truck", since there are exactly zero of these crops in this region.

New comment by nomel in "Bluesky's active user base is shrinking as its focus expands beyond the app"

nomel — Wed, 12 Aug 2026 21:41:36 +0000

Even the CEO sees there's a problem with this: https://gizmodo.com/the-ceo-of-bluesky-says-he-wants-less-of...

BlueSky isn't an experience set by the owners, it's an experience set by interacting with its people.

New comment by nomel in "A new study of a bot running a store finds it is friendly but not very smart"

nomel — Wed, 12 Aug 2026 21:19:46 +0000

I think you miss the simple point I'm trying to make: context can be a practical form of working memory.

You said

> There are no past experiences.

I'm saying that if the context contains something added in the past, it's functionally no different than a "past experience". For example, a context that is a chat log you had with the LLM. If you tell the LLM to remember a word, continue your conversation, and then tell it to recall that word, the system can recall that past experience. That "memory" that is the context feed to the LLM is a form of memory, in some cases. Now you say, "but that chat log storage that you send to the LLM is not part of the LLM!" and then I say "see above".

New comment by nomel in "Pixel Watch 5"

nomel — Wed, 12 Aug 2026 19:33:05 +0000

It would be neat to see the power budget breakdown. I suspect power draw is primarily the oled display, with the "more difficult" health sensors being second.

New comment by nomel in "A new study of a bot running a store finds it is friendly but not very smart"

nomel — Wed, 12 Aug 2026 18:58:01 +0000

Sure, but nobody uses an LLM in isolation. They use systems that are built around an LLM. In many of those systems, like a chat interface, the context is almost always used a form of memory. I think I was correct that your disagreements based on the physical location of where the previous context is stored. At the functional system level, with nearly any tool you or I use, the context/RAG/etc is used as memory, loaded with relevant/past symbols.

Say I created a new LLM architecture that integrated this concept of only appending to previous context, with no way to clear it. With it now being physically closer/integrated with the bit of code you call an "LLM", would you then agree it was memory? I don't see that implementation detail being all that important.

New comment by nomel in "A new study of a bot running a store finds it is friendly but not very smart"

nomel — Wed, 12 Aug 2026 16:36:29 +0000

LLM context window is the same concept as working memory. An LLM loading its working memory from reference a reference text or note you made is similar to how you might, if you forgot something.

Context is a type of memory, by all definitions of "memory", with the "past experiences" bit of it being the previous messages sent. You're perspective isn't clear, but if you're arguing that place this context is stored between sessions is too far away from the LLM, in a technical rather than functional sense, then I don't think that small implementation detail of a larger system (that's not even true with modern hosting) is a useful thing to critique.

New comment by nomel in "A new study of a bot running a store finds it is friendly but not very smart"

nomel — Wed, 12 Aug 2026 01:27:32 +0000

https://en.wikipedia.org/wiki/Working_memory

New comment by nomel in "$580M undersea cable rerouted to avoid the grave of Dobby the House Elf"

nomel — Wed, 12 Aug 2026 01:06:52 +0000

Actual article: https://www.theguardian.com/environment/2026/aug/10/harry-po...

Podcast where this is all from, "Can We Connect the World's Electricity Grids? with Simon Ludlam" (google it for apple/spotify/etc links)

New comment by nomel in "WorldClaw Agentic 3D open-world generation at scale"

nomel — Wed, 12 Aug 2026 00:54:23 +0000

Have you tried meshlab [1]? I've had pretty fantastic results reducing meshes from millions to some thousands.

[1] https://www.meshlab.net

New comment by nomel in "Making holograms with a pen plotter"

nomel — Tue, 11 Aug 2026 22:41:16 +0000

I guess I don't know how sharp a pick with an orange handle is from harbor freight, compared to a needle.

New comment by nomel in "To save C, we must save ABI (2022)"

nomel — Tue, 11 Aug 2026 20:54:59 +0000

Requesting a block of system memory, by address, and writing to it.

This is common with C, when interfacing with hardware.

New comment by nomel in "Go is an ideal language for AI-assisted software engineering"

nomel — Tue, 11 Aug 2026 19:38:10 +0000

I've had a really terrible time getting LLM to properly handle errors as return values. It seems that bubbling up errors, in a side channel, to a contextually relevant point in the code (exceptions) seems MUCH easier for LLM to reason about/implement properly.

Maybe my problem is I'm using a language with exceptions, so trying to go against the statistical grain, with return values, is just too much.

New comment by nomel in "Making holograms with a pen plotter"

nomel — Tue, 11 Aug 2026 19:35:20 +0000

Related, "abrasion holography", with great good success for being hand drawn: http://amasci.com/amateur/holo1.html (1995)

Maybe could swap out the pen for a needle?

New comment by nomel in "The Tragedy of the Cognitive Commons"

nomel — Mon, 10 Aug 2026 21:35:03 +0000

Seems like pretty clear breadth vs depth tradeoff, with a reduced total area.

People rarely cared about the details to begin with. Now those that did care can no longer justify the time that was previously required.

New comment by nomel in "Extreme 220GHz+Broadband Silicon Capacitor X2SC 0201M 22nF BV11"

nomel — Mon, 10 Aug 2026 21:23:10 +0000

I heard this, and have witnessed it. One of my first jobs included board rework. There were several times when I had to go ask if they wrote a uF or an nF. The response was always "obviously a uF", because nobody uses nF!