yldedly

5 ms·

1.
by yldedly 2y ago
According to the theory, an attention model. Here's artificial consciousness in three steps: 1. Have a robot build perception models of its environment and itself 2. Have the robot allocate computational resources and sensory bandwidth
2.
by yldedly 2y ago
Do have some proposal for a narrative that could work?
3.
by yldedly 2y ago
It's not a comforting story, but I think it can provide meaning. It does for me at least. While we are not gods, we are different from all other animals - we are where "the fallen angel meets the rising ape" as Terry Pratchet
4.
by yldedly 2y ago
Why do you think that is? Is this narrative too complicated, or too different from the sort of myths that we naturally tend to create? I wonder whether it'd be possible to make the story more inspirational by telling it in more accessi
5.
by yldedly 2y ago
Yes, it's correct to say that I have knowledge of your conjecture, and in the same way that physicists have knowledge of QM and GR regardless of their truth status, but beyond just having knowledge of the theory , they also have knowl
6.
by yldedly 2y ago
My point is that QM and GR make very different claims about what exists. Perhaps it's possible to unify the descriptions. But more likely there will be a new theory with a completely different description of reality. On small scales, G
7.
by yldedly 2y ago
You've called J and T into question, so let's do B as well. Physicists know that QM and relativity can't be true, so it's fair to say that they don't believe in these theories, in a naive sense at least. In general
8.
by yldedly 2y ago
Didn't know about these, thanks for the pointer! Do you have a good resource for learning about these (specifically about the hardness of sampling from posterior distributions)?
9.
by yldedly 2y ago
I think a much closer analogy to function inversion is MCMC (or Bayesian inference in general), where we can easily compute the density p(x) of any point x, but finding the x given a p(x) is intractable. Strictly speaking it's about fi
10.
by yldedly 2y ago
Even simple inference problems are NP-hard (k means for example). I think what matters is that we have decent average case performance (and sample complexity). Most people can find a pretty good solution to travelings salesman problems in 2
11.
by yldedly 2y ago
The argument in the paper (that AGI through ML is intractable because the perfect-vs-chance problem is intractable) sounds similar to the uncomputability of Solomonoff induction (and AIXI, and the no free lunch theorem). Nobody thinks AGI i
12.
by yldedly 2y ago
https://en.wikipedia.org/wiki/Paradox_of_tolerance
13.
by yldedly 2y ago
Most people in ML, even if they are very proficient, don't understand why models should generalize out of domain. They just don't think about it.
14.
by yldedly 2y ago
Fair enough, but for the sake of this conversation, if we say 'good' values are those that keep things from staying the same, aren't the values of smart people just as likely to evolve towards 'bad' ones? For exampl
15.
by yldedly 2y ago
Then what do you mean when you say "make a difference"?
16.
by yldedly 2y ago
>it's what capable and smart people value and pursue that makes all the difference. How do you know capable and smart people will keep having good values? Seems to me that it's true until it isn't - populism takes over pol
17.
by yldedly 2y ago
Do you still see such dynamics in the coefficients if you have an order of magnitude more data or fewer dimensions? 100 points in 6D is not much even for linear regression, the model might just be too high variance to interpolate monotonica
18.
by yldedly 2y ago
"Rives sees ESM3’s generation of new proteins by iterating through various sequences as analogous to evolution." Except for the part where a sequence is actually deemed more fit, ie natural selection? And the part where mutations
19.
by yldedly 3y ago
To your point, I believe AlphaStar had access to both the visual input and an API for most actions in the game.
20.
by yldedly 3y ago
I imagine that the problem with getting LLMs to do it is 1)hallucination and 2)a lack of training examples of teaching through text. We rarely teach purely through text messages, I think this might explain why LLMs almost never ask clarifyi
21.
by yldedly 3y ago
I wonder about the latter. It's known that high quality, one-on-one teaching can greatly improve learning speed. Could near future generation AI learn to be a teacher on par with the best human ones?
22.
by yldedly 3y ago
If you know how sensory organs encode information as spike trains and the many ways neurons input, process and output signals, there's no way around the hypothesis that the brain does information processing. Denying that is just trying
23.
by yldedly 3y ago
It seems that what's really needed is a formalization of the uncertainty of scientific claims, which I'm not sure is even possible to implement fully. But it's perhaps possible to schematize the relevant information - a datab
24.
by yldedly 3y ago
The Brightmind app is excellent, and you can adjust the duration of the guided meditations.
25.
by yldedly 3y ago
Probably not a solution, but many people have reported benefits from fasting, which increases cellular autophagy which degrades disfunctional cell parts - and might play a role in programmed cell death, though the relationship is not clear.
26.
by yldedly 3y ago
The projection and MLP layers don't compare all embedding pairs like attention does, so they can't distinguish between contexts where delimiters are low- vs high-importance. The projection layer mixes the multi-heads in the same w
27.
by yldedly 3y ago
Voyager is pretty cool, but it's not transferable to the real world at all. The automatic curriculum relies on lots of specific knowledge from people talking about how to get better at Minecraft. The skill library writes programs using
28.
by yldedly 3y ago
Another team beat AlphaZero with adversarial attacks: https://arxiv.org/abs/2004.09677
29.
by yldedly 3y ago
What do you mean? Attention is just matrix multiplication and softmax, it's all feed forward.
30.
No shortcuts to knowledge: AI needs to ease up on scaling and learn how to code(deoxyribose.github.io)
1 points by yldedly 4y ago|0 comments

More ›