shwaj

Makes sense, thanks. I wonder whether human web-browsing strategies are optimal for use in a LLM, e.g. given how much faster LLMs are at reading the webpages they find, compared to humans? Regardless, it does seem likely that Google’s dataset is good for something.

Can you say more about using RL at inference time, ideally with a pointer to read more about it? This doesn’t fit into my mental model, in a couple of ways. The main way is right in the name: “learning” isn’t something that happens at inference time; inference is generating results from already-trained models. Perhaps you’re conflating RL with multistage (e.g. “chain of thought”) inference? Or maybe you’re talking about feeding the result of inference-time interactions with the user back into subsequent rounds of training? I’m curious to hear more.

I love how you respond seriously to the logistical challenges of salt water, and don’t blink at the prospect of using nukes to create a tidal wave.

It’s unfortunate that Claude can’t provide references, but it plausibly suggests that the indigenous population actively managed forests in a radius around their population centers, not the entire land area. So the current larger population doesn’t make the task easier, since the size of the task scales with the population size. Superior technology should certainly make it easier, though. Probably a matter of lack of political will, and bureaucratic inefficiency.

“Filters are delays” doesn’t imply “delays are filters”. In particular, the type of effect known as a delay (e.g. a delay guitar pedal) isn’t a filter, certainly not in the musical sense, which is the relevant sense here.

Me too. White screen on Chrome, iOS and macOS.

By their definition, an amplitude envelope would probably also be a transformation.

Delays aren’t filters.

Sorry if I wasn’t clear. I meant that elite military forces make peace with death as part of their training. So, “having a lot to lose” would be less of a deterrent compared to those without such training.

The “new generation of paramilitary leaders” described by the article includes “doctors, career cops and government attorneys”. All of these would seem to have a lot to lose. If this is true, one would think that elite military operators might be even more likely to take risks for what they believe in. They’ve already made peace with the possibility of death.

Furthermore, the journalist explicitly encouraged suspicion toward Williams:

> Remarkably, the AP3er defended Williams’ loyalty each time I asserted he’d secretly tried to undermine the group.

Is this how journalists protect sources? Seems strange to me.

I agree completely, and yet I would still prefer language to be used consistently.

I know it’s not necessarily the same people, but it feels contradictory for this community to say “copyright infringement isn’t theft” when we’re talking about movies, but use the opposite language when talking about GPL source code.

AFAICT this has very little to do with the military. Quoting from the actual bill:

(A) IN GENERAL.—This Act shall not apply to classified source code or source code developed primarily for use in a national security system (as defined in section 11103 of title 40, United States Code).

(B) NATIONAL SECURITY.—An exemption from the requirements under section 3 shall apply to classified source code or source code developed—

(i) primarily for use in a national security system (as defined in section 11103 of title 40, United States Code); or

(ii) by an agency, or part of an agency, that is an element of the intelligence community (as defined in section 3(4) of the National Security Act of 1947 (50 U.S.C. 3003(4)).

Me too. At least they’re happy to capitalize “PhD”, oh wait, that somehow makes it worse.

I wonder what ways they might learn to obfuscate their thinking if they consider that their scratchpad is also open to inspection by their trainers?

That’s a different question. I was operating under the assumption (from your original post) that 1) and 3) are lies, and inferring from there that 4) would also be a lie.

That doesn’t imply that if 3) is true that it’s not a threat.

Although, haha, now that I read again, my original conclusion is wrong. If 1) and 3) are lies then that means the drones belong to the USG and they know it. Therefore that would mean that the drones really aren’t a threat. At least not to the USG.

Difficult to see how if 1 and 3 are lies, then 4 isn’t also a lie. Either they’re dealing with a threat, which is something to worry about, or they’re not but are nevertheless lying to the public, which is something to worry about.

It's possible that the secure rng incorporates observed entropy; there might not be enough entropy available when you need it, if secure is made the default and therefore used unnecessarily.

Citation needed, but of course you can’t provide those numbers. Did the republicans also have higher birth rates, and managed to grow a new batch in time to win the 2024 election?

When I’ve used AI for writing shell scripts it used a lot of syntax that I couldn’t understand. So then I took the time to ask it to walk me through the parts that I didn’t understand. This took longer than blindly pasting what it generated, but still less time than it would have using search to learn to write my own script. With search, a lot of time is spent guessing the right search term. With chat, assuming it generated a reasonable answer (I know: a big assumption!), my follow-up questions can directly reference aspects of the generated code.

What about Ergo? It’s both eUTXO and PoW.

Bar Bar Barbara Santa Barbara…

I don’t have a problem with any of that. I don’t fetishize cars quite like you describe, but I do enjoy being on a 3 hour drive because I know nothing is going to interrupt me, and usually I’m listening to music while I do.

My issue is the debasement of the language by a professional writer. This use of “categorically” is akin to “I was, like, literally starving”, and for whatever reason this grates on me. I stopped reading immediately.

“… when I sit in the driver’s seat – categorically the best place to listen to music alone“

:eye-roll: I suppose, if you don’t mind road noise. Can’t stand this overwrought writing style.

Now do it for hackers :-)

Maybe it’s possible to build something analogous to LLMs, but it’s going to need to be based on completely different technology, due to: 1) the lack of abundant training data (nothing analogous to a trillion-word internet to scrape) 2) the unacceptability of hallucinations (can’t just dose people; need some other way to validate)

They didn’t say that at all. It seems like you wanted to say it.

I’m not sure what the criteria are for “making sense”, but he’ll probably get more votes than some of the other third party candidates, despite suspending his campaign. Makes sense to me.

It’s a bit nuanced to put into a 30 second ad, but he requested his supporters to vote for Trump in battleground states and to vote for him in races that aren’t close (e.g. California).

Given his overall goal of not letting the Democrats win, this is a reasonable position for him to take. It is also reasonable for his supporters to vote for him, both to show their dissatisfaction with the other choices, but also because it sounds good for a future race if he were able to say e.g. “8% of people voted for me even though I dropped out”.