andrewla

27 ms·

451.
by andrewla 2y ago
Way weirder than this is that LLMs are frequently correct in this task. And if you forgo the counting and just ask it to list the letters it is almost always correct, even though, once again, it never sees the input characters.
452.
by andrewla 2y ago
Except that the Bitcoin only has value so long as the cryptography behind it is secure. If it is broken, then the value drops to zero and all your coins are worth nothing.
453.
by andrewla 2y ago
Thanks for this pointer -- I had read a Knuth paper ages ago that talked about this, but I couldn't remember the term of find the paper again, and this led me directly to [1] which led me back to [2] [1] https://peteroupc.gi
454.
by andrewla 2y ago
I haven't dug deeper, but the claim that the von Neumann approach is optimal does not seem intuitively correct. It seems like you could squeeze a tiny bit more entropy from it -- basically, if you reject two pairs in a row, the nature
455.
by andrewla 2y ago
Is the plan to expand from weather forecasting into climate simulation? Given the complexity of the finding initial conditions on the earth, a non-physical (or only implicitly-physical) model seems like it could offer a very promising alter
456.
by andrewla 2y ago
I love the idea of having a defer statement in general for C, but given the complexity of C scopes it's a little hard to wrangle (switch statements in particular break a lot of assumptions about how scopes should work). I would prefer
457.
by andrewla 2y ago
This seems orthogonal to the concept. More efficiently than what? Having the public key is easier than having an address because an address is the hash of a public key. So in order to crack an address, you must first find a public key that
458.
by andrewla 2y ago
Simplistically, yes, obviously true. But how does this theory explain why people lose weight on semaglutide?
459.
by andrewla 2y ago
I believe Yudkowsky has put forth a theory that somewhere in our environment there is a "GLP Supercharger" that's causing the opposite effect of the GLP-1 agonists. If a lizard bite can make you thinner, maybe there's a
460.
by andrewla 2y ago
The initial patent for semaglutide specifically was issued in 2008 and is under protection for 20 years -- while liraglutide has been around for much longer, semaglutide is relatively new in patent terms. So presumably the licensed DMF faci
461.
by andrewla 2y ago
In the end this comes down to a Reddit post from Henry's, one of the online compounding pharmacies: > There are now around ~20 FDA licensed facilities that have a DMF for the active pharmaceutical ingredient (API) Semaglutide includ
462.
by andrewla 2y ago
I thought they had an attractive agent distract him for a moment while another agent grabbed his still-unlocked-and-open laptop to prevent him from locking it or closing it up. At least I think that was the cloak-and-dagger story I heard.
463.
by andrewla 2y ago
2031 is when it becomes entirely (in the US) unencumbered by patents for its use in weight loss according to Wikipedia. The initial patents seem to date to ~2008 which means that as of 2028 the synthesis will be free of patents, but you won
464.
by andrewla 2y ago
I had a friend who worked in federal law enforcement who once described a vampire device that they used. It would clamp around a power cable and inject a UPS in the mix so that an electronic device could be removed without turning it off. S
465.
by andrewla 2y ago
If Clark had a gun to my head and Olof was refusing to give them what they wanted I would be pretty upset with Olof but way more upset with Clark. I might convince myself that Clark is super nice and only holding a gun to my head because of
466.
by andrewla 2y ago
> Not every hostage situation is actually life threatening Really? I would say that every hostage situation is tautologically life threatening -- that's what makes it a hostage situation.
467.
by andrewla 2y ago
"Told them point blank" -- I mean, there was no private channel where police could confidentially tell the hostages about this. It's a negotiation point -- make the hostage-takers aware that they do not have unlimited immunit
468.
by andrewla 2y ago
I'm glad to learn more about the incident itself, but honestly the article does not give me a real reason to doubt that the phenomenon is real. Calling it a "diagnosis" is an attempt to medicalize it, which seems ridiculous a
469.
by andrewla 2y ago
This is something that people have toyed with to improve the quality of LLM responses. Often instructing the LLM to "think about" a problem before giving the answer will greatly improve the quality of response. For example, if you
470.
by andrewla 2y ago
This is a transcription of a literal quote from the article: > Therefore, after weighing multiple factors including user experience, competitive advantage, and the option to pursue the chain of thought monitoring, we have decided not to
471.
by andrewla 2y ago
What's amazing is that given how LLMs receive input data (as tokenized streams, as other commenters have pointed out) it's remarkable that it can ever answer this question correctly.
472.
by andrewla 2y ago
The conclusion here seems improbable at best -- if I understand it right, the assumption is that somewhere in the training data is the literal token string (wild)(cat)[other tokens](w)(i)(l)(d)(c)(a)(t)? Even a transformer trained exclusive
473.
by andrewla 2y ago
So it is my contention that LLMs exhibit behavior far beyond what we could reasonable predict from a next-token-prediction task on its training set. Therefore I don't really like the framing of "this is present in the training dat
474.
by andrewla 2y ago
The lack of transflective and reflective display in common use is a mystery to me too -- while e-ink dominates on power usage and contrast, reflective and transflective displays (potentially) allow much greater pixel density and refresh rat
475.
by andrewla 2y ago
My understanding is that this uses a reflective display (like an old calculator), not a transflective display.
476.
by andrewla 2y ago
Yes! They are commonly wrong in this, and that's fascinating too. Because they are not solving the problem by looking at the letter in the word because they are not architecturally capable of enumerating the letters in the word. The fa
477.
by andrewla 2y ago
You misunderstand me -- my position is that we are completely ignorant as to how LLMs do what they do, and it is delusional to think that we understand what their capabilities are or how they are limited. Until we understand why they someti
478.
by andrewla 2y ago
That's not the sense of the metaphor that I'm applying when I say "uses language". That's closer to saying that "Alexa uses language", where "uses" here is analogous to what a calculator does. To
479.
by andrewla 2y ago
> of course it doesn't have that capability Yes, of course not. Of course. There's no way, for example, that I could ask it if it knew what number I was thinking of, and, after I tell it the number and ask the same question, th
480.
by andrewla 2y ago
> Why would a language model do anything other than "hallucinate" Indeed, why?!?!?! Why do they so often get the correct answer to very direct questions? Saying "it's in the training data" -- I dare you to find a

More ›