Ukv

47 ms·

481.
by Ukv 2y ago
What do you think "real" understanding is? To me, if something can map some modality (the word "horse", a picture of a horse, ...) into corresponding representation in its internal semantic space then I'm fine to sa
482.
by Ukv 2y ago
> Wrong! The "Large Language Model" obviously lacks a coherent understanding of language. It's not all-or-nothing for the entirety of language, but yeah it's known that LLMs are imprecise at mapping from tokens to cha
483.
by Ukv 2y ago
> It simply spits out the next "likely" occurrence I'd claim that the "likely" continuation is a bit of a nebulous concept, or at least not always trivial to work out. Completing sequences like "Q: Solve `7x
484.
by Ukv 2y ago
> that "couldn't be done" usually is in relation to some constraints or performance characteristics, which the "done" thing still does not meet I'd say theoretical proofs of impossibility tend to make valid
485.
by Ukv 2y ago
> should that math equivalence hold for programming though? It's not a hard rule that overrides all other considerations, but I do think Python's choice to follow math is the right decision here. I'd claim if you teach som
486.
by Ukv 2y ago
> the symbol it is using is overloaded with binary comparison operator, so the people will be confused I think most people would expect expressions like `5 < x < 10` to work as they do in math, without necessarily thinking about it
487.
by Ukv 2y ago
I think anti-machine-learning suffers a similar problem to anti-GMO in that animosity directed primarily towards the technology itself easily loses aim on opposing exploitative power structures, sometimes instead even achieving the opposite
488.
by Ukv 2y ago
> Would you say that telekinesis "does not seem an impossibility" as well? Why not developing engineering today that could help humanity leverage telekinesis in millions/billions of years? We are doing R&D on brain–c
489.
by Ukv 2y ago
POSIX defines a line as: > 3.185 Line > A sequence of zero or more non-<newline> characters plus a terminating <newline> character. So a file with some characters but no trailing newline is reported by `wc -l` as having ze
490.
by Ukv 2y ago
> We don't need to be able to launch 1000 Starships/year to protect the Earth from asteroids. And we definitely don't need to reuse them, it's not like we have to handle an asteroid threat every day. NASA's aster
491.
by Ukv 2y ago
> > Depends what you view as humanity's goals > Well what I quoted specifically mentions cost and ecological impact. I think those need some affirmative goal to attenuate - e.g: optimizing cost of expanding humanity's c
492.
by Ukv 2y ago
While OSI don't require the training data itself to be under a FOSS license, they do still at least require "the complete description of all data used for training", including where to obtain it, such that "a skilled per
493.
by Ukv 2y ago
Depends what you view as humanity's goals - learning more about the universe is a big one for a lot of people. In the very long term, having a good grasp on space exploration and interplanetary travel should indirectly help with other
494.
by Ukv 2y ago
> [...] there’s no way that Zuckerberg can make that look cool. The people who spend hour after hour in his metaverse will be [...] nerds and incels and the most disgruntled members of society, each desperate for escape. I think similar
495.
by Ukv 2y ago
> OSI: "data is essential for understanding and studying the system" The full quote is: > Instead, we concluded that while data is essential for understanding and studying the system, it’s not the “preferred form” for making
496.
by Ukv 2y ago
I don't think it's necessarily absurd to expect accuracy from statistical methods - in many domains (including voice transcription) they blow the accuracy of traditional non-statistical approaches out of the water, and in some are
497.
by Ukv 2y ago
I agree. For definitions, I'd roughly say: * If something changes to refine its future behavior in response to its experiences (touch hot stove, get hurt, avoid in future) beyond the immediate/direct effect (withdrawing hand) then
498.
by Ukv 2y ago
> When we ask an LLM a question, something fascinating happens. It scours through its vast training data [...] At inference time the model has no direct access to its training data (excluding RAG). Could still argue that the model's
499.
by Ukv 2y ago
Connection to AI seems tenuous. The article mostly just juxtaposes an existing scam with speculation like "An AI tool might be able to recognize a vacant property in a database faster than a human could or identify homes without mortga
500.
by Ukv 2y ago
Data I'm seeing[0] has pedestrians alone at ~35%. I think "largely" is justified, particularly when it's more than car drivers and passengers combined. [0]: https://pbs.twimg.com/media/GS2X2c4WMAAJlB
501.
by Ukv 2y ago
> This irrational hatred of personally owned cars is puzzling. The reasons don't seem that puzzling. Road traffic accidents are the cause of 1.2 million annual deaths (largely pedestrians and cyclists), the leading cause of death fo
502.
by Ukv 2y ago
A TI-84 Plus calculator (over 20 years old) is still useful today. In isolation, I don't think a model necessarily becomes less useful over time. It'll still be as good at summarizing articles, translating text, correcting grammar
503.
by Ukv 2y ago
> https://edintegrity.biomedcentral.com/articles/10.1007/s4097... > GPTZero was correct in most scenarios where they used basic prompts, and only had one false positive. One false positive out of only "
504.
by Ukv 2y ago
> [...] but a viable model where students and laypeople have access to it? To be seen. You can run GPT-4-equivalent models locally. Even if all software and hardware advancements immediately halt, models at the current level will remain
505.
by Ukv 2y ago
> Such as? We need an answer now because students are being assessed now . Return to the old "viva voce" exam? Still used for PhDs. But that doesn't scale at all. For a solution "now" to the cheating problem,
506.
by Ukv 2y ago
> The claim of a new form of “dark” matter—unseen and unseeable and present in far greater quantities than the visible matter of the universe—was not a simple explanation, but it turned out to be the best explanation. I feel the key here
507.
by Ukv 2y ago
My comment is questioning JohnFen's definition that a photo must be an accurate representation of the light that was received, to be clear. Any existing photo will diverge from that in various ways, often because it better captures r
508.
by Ukv 2y ago
Sure, could say that the lens itself is altering the light and thus no longer an accurate representation of the light received by the camera. Particuarly evident with something like a fish-eye lens.
509.
by Ukv 2y ago
> [...] the light that was received [...] > Anything that makes the image vary from reality [...] I'd argue these are not equivalent definitions. Demosaicing, red eye removal, exposure/brightness/contrast adjustments, w
510.
by Ukv 2y ago
> [...] the poor treatment of various marginalized groups, etc. > [...] but if you're paying attention, and paid attention in the past, it's painfully clear that things are worse now. Treatment of marginalized groups is wo

More ›