gaganyaan

7 ms·

61.
by gaganyaan 2y ago
AGI exists now. We need better terminology. You want something more, and that's fine because I do too. Our state-of-the-art models are baby AGI, because they can accomplish a wide variety of tasks without being specifically trained for
62.
by gaganyaan 2y ago
They can be trivially shown to reason. It's quite normal for many things to have reasoning capacity, including amoeba, sunflowers, cats, LLMs, and humans. The real issue is establishing the limitations of their reasoning capabilities,
63.
by gaganyaan 2y ago
Sure, waves exist, whatever. Not to be rude, but so what? This kind of comes across as "quantum waves therefore the Universe Is Love" or something. What about the waves matters here? I'm not arguing that consciousness doesn&#
64.
by gaganyaan 2y ago
TBH, I don't really know what you're driving at with brain waves. They're not magic. I think you're trying to use "consciousness" to mean "high level brain activity", as opposed to something like the
65.
by gaganyaan 2y ago
Much of the confusion here is from terms that you use like "become conscious", which binarizes it. When run, that program will have a very simple model that includes only a few things like stdout. More complex programs will have m
66.
by gaganyaan 2y ago
Why would a perfect simulation of a brain not have brain waves? They're not magic. I'm perfectly happy to talk about brain waves. I suspect anyone else in the "Consciousness is a Computation" camp is happy to as well, I
67.
by gaganyaan 2y ago
See my other comment too, but I agree. Rocks and lightning aren't conscious of anything, because they have no state that represents anything. Amoeba are conscious of some things, because they have state that represents a tiny slice of
68.
by gaganyaan 2y ago
It's not the silicon that's conscious of things, it's the model that happens to use that silicon.
69.
by gaganyaan 2y ago
Qualia is a silly god of the gaps style argument. There's no difference between "seeing red" and "having your neurons tickled so that you think you're seeing red". People really don't understand consciousn
70.
by gaganyaan 2y ago
The LLM sees tokens, and predicts next tokens. These tokens encode a vast world, as experienced by humans and communicated through written language. The LLM is seeing the world, but through a peephole. This is pretty neat. The peephole will
71.
by gaganyaan 2y ago
Another one is to just ask it to spell it out with dashes in between each letter. If anyone's wondering, it then gets it right because that changes the tokenization, changing how the model actually "sees" it.
72.
by gaganyaan 2y ago
This isn't about neurodivergence. Every single human suffers from a long list of cognitive shortcomings. We just call them things like "optical illusions" and find them interesting, but don't then go on to make silly cla
73.
by gaganyaan 2y ago
There's as many definitions as there are dictionaries, but let's try picking "awareness". LLMs are aware of many things, so they are conscious of those things, meaning they display some level of consciousness.
74.
by gaganyaan 2y ago
Can a cat reason? It knows nothing of FOL or math, but can be very smart when figuring out how to catch a mouse.
75.
by gaganyaan 2y ago
Define consciousness before making that claim either.
76.
by gaganyaan 2y ago
Humans have many well-documented cognitive shortcomings and failure modes. Does that mean we don't reason either?
77.
by gaganyaan 2y ago
They very much do understand words, letters, etc. Why do you think otherwise? Because of one trick question?
78.
by gaganyaan 2y ago
Define reason before making grandiose claims.
79.
by gaganyaan 2y ago
What compromise is to be found with the parents in the first example? If you don't completely agree with them, then you're a sinner. I don't think the daughter can find a compromise.
80.
by gaganyaan 2y ago
That likely means something like "This is using a rare value for the key 'parking'. It might be incorrect, check that it's not supposed to be a more common value". In other words, it's trying to catch things li
81.
by gaganyaan 2y ago
They probably mean the poem at the bottom here: https://www.python.org/downloads/release/python-3120/ Under "And now for something completely different"
82.
by gaganyaan 2y ago
It will just end up like Qualified Immunity. "Your honor, I was wearing a green shirt, this is unprecedented territory and totally different and Congress needs to clarify". What historical uses of the Chevron deference do you ac
83.
by gaganyaan 2y ago
Because this often ends up with people talking past each other, never until a few years ago did anyone use singular they for a known, specific person. Shakespeare used it for unknown or nonspecific people.
84.
by gaganyaan 2y ago
Somebody went one step further and invented a format that uses the unicode codepoints as separators: https://github.com/SixArm/usv
85.
by gaganyaan 2y ago
"Truth" derives from experience
86.
by gaganyaan 2y ago
In the context of LLMs, that is truth, as anyone would understand it. Humans are also sensitive to recurring or consistent input in our training data.
87.
by gaganyaan 2y ago
Then keep it private, and nobody else can do anything with it. Otherwise too bad
88.
by gaganyaan 2y ago
It's odd that the article didn't mention nix, at least to say "We didn't like it for this reason".
89.
by gaganyaan 2y ago
We can invent post hoc rationalizations, just like LLMs can
90.
by gaganyaan 2y ago
Humans are infamously bad at being consistent: https://en.m.wikipedia.org/wiki/Compartmentalization_(psycho... The biggest mistake I see people make when criticizing LLMs is that they take the best possible modes of hu

More ›