gf000

15 ms·

391.
by gf000 7mo ago
You are right, but at the same time the human brain does way more stuff (muscle coordination, smell, touch sensing) and all those others take up at least some budget. So interesting question, but I'm not convinced it's only a scal
392.
by gf000 7mo ago
How does alphago come into picture? It works in a completely different way all together. I'm not saying that LLMs can't solve new-ish problems, not part of the training data, but they sure as hell not got some Apple-specific libra
393.
by gf000 7mo ago
You may want to watch this video: https://youtu.be/e7wFotDKEF4?is=bl5TPvk9_mdnG3Om Human language is way above what communication animals show. We don't really know what's the exact boundary, but again, the diff
394.
by gf000 7mo ago
Humans are "multi-modal". Sure we get plenty of non-textual information, but LLMs were trained on basically every human-written world ever. They definitely see many orders of magnitude more language than any human has ever seen. A
395.
by gf000 7mo ago
How is that relevant? The human brain is at the point of birth (or some time before that). We compare that with an LLM model doing inference. The training part is irrelevant, the same way the human brains' evolution is.
396.
by gf000 7mo ago
AGI is definitely possible - there is nothing fundamentally different in the human brain that would surpass a Turing machine's computational power (unless you believe in some higher powers, etc). We are just meat-computers. But at the
397.
by gf000 7mo ago
https://xkcd.com/605/
398.
by gf000 7mo ago
How confident are you that this knowledge was not part of the training data? Was there no stackoverflow questions/replies with it, no tech forum posts, private knowledge bases, etc? Not trying to diminish its results, just one should a
399.
by gf000 7mo ago
I would argue that point. The biological components are the same, but emergent behavior is a thing. So both the scale and the number of connections/way they connect have surpassed some limit after which cognitive capabilities increased
400.
by gf000 7mo ago
The costs keep decreasing and self-hosted models may be able to do some of the tasks as well. So this may be only temporarily unavailable for many.
401.
by gf000 7mo ago
https://arxiv.org/html/2502.16763v2 You are wrong. Especially that we are talking about models with 50T parameters. Can they do arbitrary computations for arbitrarily long numbers? Nope. But that's not remotely th
402.
by gf000 7mo ago
With all due respect, this is just plain false. The reason "strawberry" is hard for LLMs is that it sees $str-$aw-$berry, 3 identifiers it can't see into. Can you write down a random word your just heard in a language you don
403.
by gf000 7mo ago
> Really ? Every Human ? Yes, in many ways absolutely. Just because a model is a better "Google" than my dummy friend doesn't mean that this same friend is more capable at countless cases. > Meaningless comparison. You
404.
by gf000 7mo ago
Humans are obviously unique in an interesting way. People only "move the goalpost" because it's not an interesting question that humans can do some great stuff, the interesting question is where the boundary is. (Whether agai
405.
by gf000 7mo ago
I think it's more of a data vs intelligence thing. They are separate dimensions. There are problems that don't require any data, just "thinking" (many parts of math sit here), and there are others where data is the signi
406.
by gf000 7mo ago
So you don't think 50T parameter neural networks can encode the logic for adding two n-bit integers for reasonably sized integers? That would be pretty sad.
407.
by gf000 7mo ago
What calculations? Do you mean "3+5" or a generic Turing-machine like model? In either case, this "it's a language model" is a pretty dumb argument to make. You may want to reason about the fundamental architectur
408.
by gf000 7mo ago
> a problem that had already been solved by any other desktop operating system in the last 50 years I just gave you an example of Windows that by default fails this requirement (see settings vs control panel or what that is called), let
409.
by gf000 7mo ago
> there are no solutions, only tradeoffs Sure, but we are talking about desktop OSs which do specify a set of requirements to fulfill, and in that scenario I do believe both systemd and especially nix brings a whole lot of benefits at th
410.
by gf000 7mo ago
> That because a similar things happen fairly regularly in Linux land. It seems every 5 years (or so it seems) There are things that are working perfectly fine, and then someone/some group decided that half of a particular stack nee
411.
by gf000 7mo ago
That's what stuff like SDL meant to abstract away.
412.
by gf000 7mo ago
Well, bloated just got a whole new meaning! Hopefully the gases are temporary only!
413.
by gf000 7mo ago
Frankly I don't get your problem or how is it different on any other OS. So your solution to GTK or qt looking alien is to look alien to everyone? Like there is no universe where "GTK doesn't look good, I will go with a custo
414.
by gf000 7mo ago
Afaik pijul already does that though
415.
by gf000 7mo ago
And why should we optimize for "how many locs can I write a screenshot tool in"? I would prefer to having 1-2 decent ones that have all the necessary features and call it a day. They can then have a simple API so you can script wh
416.
by gf000 7mo ago
Why would it be more complicated? Most apps are just using GTK and qt and doesn't even care about their x or Wayland backends.
417.
by gf000 7mo ago
It's only using sockets to pass a handle to a buffer though. Sockets are plenty fast, but not moving stuff in the first place can't be beaten.
418.
by gf000 7mo ago
It never worked very well - in the age of bitmaps (and that's not a recent invention) that ultra-dumb pixel pushing simply no longer scales. We have eons better ways to transport graphical buffers around (also, what about sound? That s
419.
by gf000 7mo ago
Are they not just using libinput and that's it?
420.
by gf000 7mo ago
> Are you arguing that the Android Ecosystem uses Java Now you are just arguing in bad faith. Who talks about Android, which historically lagged a decade behind OpenJDK and not particularly good at being up to date even today, so people

More ›