jksk61

Is a TL;DR available or at least some of the ideas covered? Because after 3 paragraphs it seems the good old "it is actually something resembling a cellular automata" post by Wolfram.

funny paper, I still don't know what was the goal of it. It is evident to anyone that LLM can't perform any meaningful reasoning, why even bothering in building such an infrastructure to test whether it is able to become a "scientist".

not even childish gambino ref saved it.

in fact it is nothing new.

"you were supposed to destroy evangelisms not join them!"

now the question is: what is the most complex* object that it is not turing complete?

* let's say you have n distinct rules acting against a set S, the complexity is n.

p.s. probably something trivial exists such that you can take n as large as we wish to, so probably my definitions are not interesting.

bruh! with 100 mln elements in the dataset it is basically printing out previously known problems during testing.

I don't get it, what's cool about it? Einstein notation seems good enough for most of the things and they are equivalent. Is there anything interesting (i.e. new) that this notation allow?

> What's the strength of play for the GPT architecture? It's impressive that it figures out the rules, but does it play strong chess?

sometimes it is not a matter of "is it better? is it larger? is it more efficient?", but just a question.

mountains are mountains, men are men.

it seems to try too hard on proving things have improved by white lying a bit.

Some examples:

* Environment: air quality in most places has continued to improve (and considering the growing evidence on the harms of air pollution, this may well be the single most important item on this whole page), forest area has increased, and more rivers are safe to fish in - yeah except that are only statistics related to USA, not to mention the Detroit area water pollution thing that it is still going on. Or all the PFAS related drama happening recently.

* LASIK surgery has gone from an expensive questionable novelty to a cheap, routine, safe cosmetic surgery - yeah, almost cheap, safe sure, but none of the people I met actually says they'll do it again (sample N=5), for various reasons. Some: they did have to take glasses again after a few years, colors were less vivid, eyes were less hydrated.

* Food, it is nice that fast food aren't shitty anymore, but there are still plenty of contaminations happening (not only in USA but all-over the world). Moreover, the food is becoming worse because of the lack of microsubstances (and soil degradation) as well as the increase in CO2. Agriculture output is damped by climate change.

very nice!

If you need an exponential number of coupled oscillators to construct a general purpose computer, then you don't have the exponential speedup.

of course not, none is considering fusion as a reliable workable solution for the 2050 net zero goals. I mean they don't even consider opting out gas and oil until 2040s or even later lol

I don't know about you but I use chat gpt to solve captcha because I can't.

yes it is an interesting read, however you should know that some of their "facts" are highly speculative and some were proven wrong in the last decade.

color me surprised.

also removing skip connections leads to a rougher loss landscape, hence it should be harder to find the optimal weights.

I don't know, it is an opinion and it seems kind of well-founded (i.e. there's no evidence for groundbreaking research on OpenAI part except for scaling things up).

I don't understand why investing money in something that it is reliable and not requiring major reworks on the electric grid would be beneficial, incredible! I am more baffled than you.

also, in case you didn't know sweden doesn't have a huge amount of sun (especially in winter), wind is more affordable but you can't rely on a unique energy source.

what a laughable sentence, there have been at least 2 near misses last year between starlink satellites and Tiangong. Also, if there's a collision, the smaller parts and debris would remain in orbit far longer than expected (hence causing more incidents).

these kind of books are "philosophical" books. In the sense that the usage can be written/explained in other ways. For instance, monads are just a way to see that if you compose two ducks you still have ducks that behave in the same manner, but that's almost the same thing as having a class and define the composition arrow as a method.

The only difference is that with CAT you have theorems asserting what's true and what's not, with classes I don't think there's much literature. A similar reasoning can be applied to neural networks as there are very different POV: someone will look at them as ODE systems, other as a pure discrete dynamical system or as an optimization problem or whatnot.

well it could be useful in literature where there are way too many zhang...

Nice, but what are the minimum requirements to run it locally?

No it is obviously Paris, Texas

one could argue that the most important criteria is that if you take any interval between 0 and 1 you get the same density... at least that would be a uniform distribution, since even a normal, exponential or whatever could have mean 0.5 and simulate that instead.

If you know the enemy and know yourself, you need not fear the result of a hundred battles. If you know yourself but not the enemy, for every victory gained you will also suffer a defeat. If you know neither the enemy nor yourself, you will succumb in every battle.

Sun Tzu.

How do you think you can devise a vaccine if you don't know the virus? They experiment to find out what are the weaknesses of the enemy (i.e. proteins and other markers), so that they can target it. If the next pandemy is caused by a similar strain: ta-dam you have the vaccine already (or at least you know what to target).

Yeah, but what about "On april 19th, I made bread"?

See also https://pastebin.com/tKLSWjSt, which is a collection of Pompei graffiti.

you are using a browser which has zero day vulnerabilities nonetheless.

Literally never needed to copy text from smartphone to pc (or viceversa). At best anyone needs to sync between photos, docs or stuff like that. For that you just need a sync on the cloud like Dropbox, Drive, Box, and so on (also never heard of bluetooth for small files?)

moreover, apple's software is generally very buggy.

dunno, seems comparable to chromium-based alternatives in terms of power drain.