Hacker News: Theory42

New comment by Theory42 in "Write Like It's 1866: LLMs Relearn Telegraphese"

Theory42 — Wed, 07 Oct 2026 15:36:01 +0000

Yeah! LLM compression is a spectrum between 'normal stuff we can read' and vectors. Depending on trust in the models and the desire for compression, there's a choice to be made on which formats you want to allow. Pretty interesting stuff.

New comment by Theory42 in "Write Like It's 1866: LLMs Relearn Telegraphese"

Theory42 — Wed, 07 Oct 2026 14:32:32 +0000

A good point. I initially had a model for a judge, but it seemed to give very lenient scores. I'm open to learning about how best to benchmark the phenomenon, though.

New comment by Theory42 in "Write Like It's 1866: LLMs Relearn Telegraphese"

Theory42 — Wed, 07 Oct 2026 14:30:45 +0000

I wrote it with the help of an LLM, then edited it, then wrote some more, then edited that. Took about a week to format. It contains my thoughts. Brave new world, I know. I'm finding it ironic that the same crowd that is merging AI code all day is so allergic to AI assistance in prose...do all your work using this new godlike technology, but when it comes time to share, whip out your fountain pen or else.

New comment by Theory42 in "Write Like It's 1866: LLMs Relearn Telegraphese"

Theory42 — Wed, 07 Oct 2026 14:17:21 +0000

A useful critique, thanks.

I would say that the grader has the same threshold for whatever answer it receives, and is equally harsh on whichever it grades. Any scores above the baseline (1.0) are really claims about parity, rather than better understanding in the compressed format.

The decoder step is a model expanding the cablese to regular text, not having seen the initial question. A separate model instance then reads that regular text and answers. And the result is still at parity with the plaintext record.

Had cablese knocked out information, that wouldn't have been the result, would it?

New comment by Theory42 in "Write Like It's 1866: LLMs Relearn Telegraphese"

Theory42 — Wed, 07 Oct 2026 13:29:09 +0000

I use the tools I study, nothing more or less.

New comment by Theory42 in "Write Like It's 1866: LLMs Relearn Telegraphese"

Theory42 — Wed, 07 Oct 2026 13:14:33 +0000

Cool story bro. Maybe you could engage with the content? I'm an actual person.

New comment by Theory42 in "Write Like It's 1866: LLMs Relearn Telegraphese"

Theory42 — Wed, 07 Oct 2026 12:52:56 +0000

I suspect you might be right. My conclusions from the digging are that this kind of compression works best with settled instructions/data for machine to machine talk. For something like OpenClaw (which I use a lot), that might mean the AGENTS.md, TOOLS.md, etc. Compression there would free up the context for the agent.

New comment by Theory42 in "Write Like It's 1866: LLMs Relearn Telegraphese"

Theory42 — Wed, 07 Oct 2026 12:36:26 +0000

Not quite, and its addressed in the post. Caveman was cool, but hand wavy. I've measured the compression of Cablese, determined where it does the most good, and also that it is already baked in to most model family's training, which saves instruction tokens.

Write Like It's 1866: LLMs Relearn Telegraphese

Theory42 — Wed, 07 Oct 2026 12:04:36 +0000

Article URL: https://fiveminutesforward.com/post/2026-10-04-telegraph-test/

Comments URL: https://news.ycombinator.com/item?id=49991580

Points: 84

# Comments: 52

New comment by Theory42 in "Japanese used bookstores see 5x sales surge as books are being bought by the ton"

Theory42 — Mon, 28 Sep 2026 13:20:44 +0000

You'd think they could give the scans to archive.org or similar. Even if the books are in copyright, they won't be forever. We can't let our culture be destroyed like this.

New comment by Theory42 in "ChatGPT Atlas"

Theory42 — Fri, 24 Oct 2025 15:18:04 +0000

Detecting genuine human angst in the sea of AI posts

New comment by Theory42 in "Federal workers ordered to return to offices without desks, Wi-Fi and lights"

Theory42 — Wed, 05 Mar 2025 21:03:22 +0000

Say it ain't so

New comment by Theory42 in "38C3: Illegal Instructions"

Theory42 — Fri, 03 Jan 2025 12:48:08 +0000

I really liked this talk as well. One part that I'm not sure I can fully agree with, though, is the idea of fully re-conceptualizing the self. It is possible to self-author, and partially change...but I have never heard of nor met anyone who just became a totally different person. I'm willing to concede that he may have been speaking hypothetically, or that maybe the idea of changing the self will be more accessible to AGI rather than for humanity.

New comment by Theory42 in "The Albuquerque "Broken Arrow" Nuclear Accident"

Theory42 — Fri, 20 Dec 2024 17:00:07 +0000

If the plutonium core was removed, as indicated in the blog, why was the impact crater radioactive?

New comment by Theory42 in "Proton is taking its privacy-first apps to a nonprofit foundation model"

Theory42 — Sat, 22 Jun 2024 10:16:10 +0000

Why aren't cooperative business models more common among software companies? Starting a software company typically doesn't require a large amount of capital, especially with open source projects where individual contributions can play a significant role. In these cases, a cooperative model could effectively distribute ownership among the contributors, essentially making them co-owners. This approach seems simpler than managing a hybrid of nonprofit and for-profit elements.

New comment by Theory42 in "The Basque Country’s Mondragón Corporation is the largest industrial co-op"

Theory42 — Sat, 27 Apr 2024 13:23:12 +0000

You say that like its a bad thing...

New comment by Theory42 in "Star Trek: Borg – Remastered"

Theory42 — Sun, 25 Feb 2024 10:40:51 +0000

This is great, thanks for making the effort!

New comment by Theory42 in "Mark Zuckerberg’s new goal is creating artificial general intelligence"

Theory42 — Fri, 19 Jan 2024 18:18:28 +0000

That's the thing about having lots of money: all your bad ideas can get a workforce behind them. Admittedly, the good ones as well.

New comment by Theory42 in "OpenAI deletes ban on using ChatGPT for "military and warfare""

Theory42 — Fri, 19 Jan 2024 18:15:01 +0000

It seems you have it out for the US and Israel specifically, which is certainly a take, but not a very well rounded one. If you were, say, against the use of this particular technology for any military, that would be one thing, but you seem to only want the US DoD to not have it.

New comment by Theory42 in "LK-99: The live online race for a room-temperature superconductor"

Theory42 — Tue, 01 Aug 2023 18:12:15 +0000

All the cool swords seem to come from societies without access to good iron. Ever seen a 'Viking' sword?