abecedarius

What stands out to me: the #1 (FL) over 3x the inflow of the #2 (TX). Both are large southern coastal states with no state income tax. What's Florida got that's that much more attractive?

The usual meaning: say you have an interpretability signal that reveals reward hacking or other badness. Your most obvious response is to use that signal as a penalty in training. Doing that is called "the Most Forbidden Technique" because you can expect it to degrade the signal without stopping the badness.

But this doesn't seem to describe the OP, it's a different way to lose monitorability.

Energy cost may be a pointless comparison since we don't eat electricity, but it's very reasonable to look at sample efficiency and ask what we have there that our theories miss.

I don't understand where you're getting "hates China" unless that's a framing of "disapproves of the Chinese government on the whole".

I have seen Zvi pushing back against the narratives that open-weights models and Chinese ICs have caught up to the frontier. Was he wrong? But I can't recall any animus against Chinese people or their talent there.

Do you have inside info about that inspiration? At one point a human character monitoring from space thinks Underhill seems like several Dawn Age geniuses rolled into one and thinks he must be a figurehead for a collective (iirc). But that's a bit different from the author taking off mainly from him. vN isn't mainly known as a font of crazy ideas -- more like supremely fast & precise intellect.

It's true unless we let freedom of speech and the press be interpreted narrowly, as the right to flap our jaws and to press paper against ink. That is up to us collectively.

Try deepl.com? (But I haven't needed it in a while, maybe the suck fairy has been to it too.)

The right basic idea is, the UI for "do X to Y" must combine designating which Y you want and giving permission to access that Y (and no other). It's when the OS shell separates these aspects that you have to choose between extra annoying useless permission pop-ups and any security boundaries.

(Yes reworking our all our systems in terms of capability security is a giant job. But there's a difference between a giant job and )

No idea about forums, but for a book with more depth on these issues I liked Uri Alon's Systems medicine.

Go back five years on HN and look at what the mass of commenters considered sci-fi delusional thinking about AI. (GitHub Copilot was released just under 5 years ago, for instance, provoking some discussion of what the tech might mean for the programming job market.) Why do you think dismissal as sci-fi is better grounded now? Do you have arguments instead of dismissals?

I felt the same way.

There was another downside, that the Applesoft BASIC license gave Microsoft a hostage which they later used to kill https://en.wikipedia.org/wiki/MacBASIC. Hard to tell how much it mattered that the Mac came with no user programmability, which had been expected for PCs till then.

Is there "benchmark data" on different memory techniques? I also wondered about this because I also memorized 50 digits on a lark back in high school, without the post's encoding technique and memory palace.

I added one block of 10 digits per night, spending maybe 10 minutes per night? Very roughly an hour total, ~same as the poster spent to get 100 digits, but without the prior investment they spent on the PAO (memorizing 300 key-value pairs, apparently).

You'd want to compare to Anki, too.

The Austin main library has a large cafe on the ground floor. Good combo, though I liked the pre-covid one better. Maybe there should be two!

Fair enough, but the abstract really is too opaque imo.

> The k-server conjecture states that a deterministic online algorithm can achieve competitive ratio k on every metric space.

I was like "algorithm for what?" when I knew what every one of those terms meant. Lots of things use k servers. Lots of problems involve metric spaces.

Yes! There's both the daunting problem of technically how can we even do this, and the broader problems of what's good/acceptable and how do we resolve that among each other.

I believe this mismatch of rates of progress means we need to stop slamming the accelerator on capabilities for now even though as a libertarian I'm sure whatever governance process we manage to get to will be, uh... suboptimal.

I agree with most of this, but you're misunderstanding "alignment" as coined. Yes, training powerful enough AI, any simple optimization target gets you malign behavior, because human values are not simple. If you insist on making powerful AI, you'd better instill respect for human values! That's "alignment".

https://www.lesswrong.com/posts/ZxWzCGKzX84S7DBZ9/when-was-t...

Was [2] meant to be https://www.amazon.com/Lines-Curves-Practical-Geometry-Handb... ? In my browser your link brought up a page with no book. I found the above by searching for Gutenmacher.

fwiw I also observed a persisting slowdown soon after upgrading my Studio to OSX 26. Since I haven't been using it very much I'm short on data bearing on the cause -- maybe it was GPU use though I don't recall doing anything like that.

Thanks! You're better at search than me. It was part of a story, described here (last 3 pages): https://www.researchgate.net/publication/403044354_Photograp...

> processes better regarded as generalization and prediction, which seem not only seem to have direct and obvious neural realizations

If the workings of those circuits are obvious to you, I'd really like to learn. Do you mean the level of analysis at https://transformer-circuits.pub/ ? (That looks like good work but not a deep understanding.)

Hofstadter referenced this back in the day as a promising beginning: https://en.wikipedia.org/wiki/Sparse_distributed_memory which sounds kind of similar in style to the embeddings you bring up.

This reminded me of a story/vignette which my fallible memory tags as by someone like Twain or Poe, about an automatic rewriting machine. It was put to work on the same passage with an ever higher intensity setting, becoming more and more gothic to the point of parody -- iirc the final output was verse including the line "to atoms dash the doubly damned".

Does anyone remember this?

Sounds like a problem that'd hit anyone running LLMs. I haven't tried on mine so far, but people do talk about ordering a $10k Mac Studio just for that. Anyone else see this? Does the OS version matter?

Author also known for inventing (ro)BDDs later in the 80s.

"The BASIC they did which we got ahold of the BASIC source code and enhanced t quite a bit was great."

Thanks! I think this is still talking about his high school volunteer hacking rather than the MS Basic era. (i.e. doesn't obviously support your "ported it".)

From [2]:

> Gates said: "I'd skip out on athletics and go down to this computer center. We were moving ahead very rapidly: Basic, FORTRAN, LISP, PDP-10 machine language, digging out the operating system listings from the trash and studying those."

Is there more context clarifying this means they studied the source code of DEC Basic?

I haven't tried this but sounds like https://github.com/cloudflare/cloudflare-os ?

This might come down to details of how you explain it: iirc Apostol took those basic moves (axioms) and calculated what the formula would have to be, rather than starting with a formula and checking that it has the properties of a signed volume.

But I don't know, it's been a very long time for me. It's good to have a variety of approaches to the subject.

It seems to me more like determinants are often badly motivated. The way I remember it from Apostol's Calculus was like "a volume multiplier would be very useful; it needs to be a signed volume for linearity, which implies antisymmetry. Here are axioms collecting these requirements. They're uniquely satisfied by the determinant. Proof: ..."

Agreed that it should help if you got to learn wedge products first (I didn't).

I didn't carefully read either of these, but it reminds me of https://blog.cloudflare.com/code-mode-mcp/ (note it predates the filing).