npostavs

But this is externally checkable: if watermarking is as described, then the rate of next-token agreement on identical prefixes should match within an account and across accounts.

With features like memory and cross chat history, getting identical prefixes across accounts might be difficult. And isn't username part of the context too?

Hmm, I just tried with ChatGPT 5.6 Luna "Generate a D&D character for me.", and got:

1. Kael Thorn half-elf ranger
2. Kael Emberstep tiefling rogue
3. Kael Veyrin, tiefling warlock

Claude Sonnet 5 seems to like half-elf rangers more, although has slightly better first name diversity:

1. Kestrel Ashwood Half-Elf Ranger
2. Kaelin Duskwarden Half-Elf Ranger
3. Kael Ashgrove Half-Elf Ranger

I don't see sufficient reason to believe this or other common LLM names like Elara Voss are due to watermarking in particular. Isn't it more that LLMs are just generally bad at maki... (read more)

If anyone would like to point to specifics I’d be grateful

The use of "genuinely" in the opening (and a few more times) definitely had me wondering.

FWIW 10-20 years earlier [...] You would have simply had no way to find your way to Philadelphia, and would have either waited or turned around and gone home.

What about paper maps? Asking someone for directions?

This is a dup of https://www.lesswrong.com/posts/jW4TeNZhxBA9Fzdim/housing-roundup-15-the-war-against-renters

This is a duplicate of https://www.lesswrong.com/posts/eSwdsDTnqigQJPfkw/housing-roundup-13-more-dakka

There are ~4.3 trillion IPv4 addresses

[...]

Assuming residential ISP connections also use CGNAT, it is possible we never (?) need ipv6 as all webservers and all clients can be serviced on 4B * 1000 IP addresses (??)

[...]

There are 2^32 = ~4.3 trillion IPv4 addresses.

2^32 = ~4.3 billion, not trillion.

I didn't vote, but it could be because it's formatted as a wall of text, seems to have a lot of useless filler and no clear message. I got to the end of it, but don't really have any idea of what you wanted to say.

Ostensibly-neutral (but in reality heavily blue-tribe-coded) "experts" who stay in their lane are not exactly popular or influential at the moment,

Aren't they at a low point due to not staying in their lane by smuggling in political views into their ostensibly-neutral pronouncements? E.g., doctors saying it's okay to break Covid restrictions on gathering if it's for social justice.

See https://x.com/twocents/status/2020596821228388704

It sounded kind of... rehearsed? Not sure if I should take this is as real position.

Hard RSI: AI modifies itself in a way that is different from just changing numerical values of its weights. It creates a new version of itself [...]

In hard RSI there is no danger of misalignment since AI doesn't create a successor, but rather modifies itself. In easy RSI there is danger of misalignment, [...]

I don't think I understand how "creates a new version of itself" is different from "create a successor"?

Oh, LLMs also suggested SCP-3125, but I thought they were wrong because "U" didn't seem like a plausible typo for "SCP". I wasn't aware of the alternate U-3125 naming.

But rebelling against a globalized techno-political-memeplex is like rebelling against U-3125.

Is "U-3125" referencing something?

I think I'm mostly following now, but when you write stuff like:

In the higher education system, I expect it would take the form of increasing the swathe of universities which taught a complete curriculum, as well as evening out the distribution of staff.

I wonder, is the undergraduate curriculum really significantly different between top-tier universities and others? Instead of wasting space on the rocket analogy, it would be useful to establish that sort of thing about the actual subject. And generally, the posts is really missing a lot of detail about... (read more)

You are talking so much about rockets that I can't even tell what point you're trying to make about universities. The post would probably be a lot clearer without this analogy.

This is an accidental double post of https://www.lesswrong.com/posts/FJxc4Lk6mijiFiPp2/the-big-nonprofits-post-2025 (also double posted on the wordpress site: https://thezvi.wordpress.com/2025/11/26/the-big-nonprofits-post-2025/ and https://thezvi.wordpress.com/2025/11/27/the-big-nonprofits-post-2025-2/)

Seems understandable to me (although I guess I'm somewhat primed by reading the previous versions).

I think most of "you" can be omitted in English as well:

Imagine: you study an immature AI in depth. Decode its mind entirely. Develop a great theory of how it works. Validate this theory on a bunch of examples. Use that theory to predict how the AI’s mind will change as it ascends to superintelligence and gains (for the first time) the very real option of grabbing the world for itself. Even then, you are, fundamentally, using a new and untested scientific theory to predict the results of an experiment that has not yet run, about what the AI will do when it really, actually, for real has the opportunity to grab power from the humans.

This seems to be an accidental repost of https://www.lesswrong.com/posts/9TPEjLH7giv7PuHdc/crime-and-punishment-1 from April. (It's also reposted on https://thezvi.wordpress.com/2025/11/03/crime-and-punishment-1-2/, but not thezvi.substack.com/).

Seems to be a repost of this one: https://www.lesswrong.com/posts/wcp8hcQwqhd8nDA2x/ai-131-part-2-various-misaligned-things (https://thezvi.substack.com/p/ai-131-part-2-various-misaligned)