Buttons840

87 ms·

1501.
by Buttons840 4y ago
I've discussed this before[0], dragonwriter has a good response explaining that this law is not automatically illegal just because it affects interstate commerce. I do hope you're right though, and that these laws are ruled uncons
1502.
by Buttons840 4y ago
Multiple States have already passed laws requiring everyone to upload their government papers (drivers license or passport, etc) to use social media. Things seem to be changing, that's for sure -- and fast.
1503.
by Buttons840 4y ago
True. I've read a lot about reinforcement learning and the exploration vs exploitation tradeoff is both technically interesting and philosophically interesting. I've also seen algorithms I've implemented fall to all sorts of
1504.
by Buttons840 4y ago
This has been reinforcement learning for decades. People say "OMG! The agent is acting and learning and getting better all on its own!" And they're right, in the beginning. Eventually the agent plateaus or collapses. Supervis
1505.
by Buttons840 4y ago
I've been reading this paper with pseudocode for various transformers and finding it helfpul: https://arxiv.org/abs/2207.09238 "This document aims to be a self-contained, mathematically precise overview of tr
1506.
by Buttons840 4y ago
In my machine learning experience, if it only takes 10x the parameters brings a significant improvement I feel lucky.
1507.
by Buttons840 4y ago
I've had a related idea. I'd like it if political debates allowed the candidates to talk to each other directly more, and I want a timer above each candidate that shows how long they have been talking and an interruption counter.
1508.
by Buttons840 4y ago
Utah passed a law that social media sites have to verify the age of users using drivers license or passport, etc. Check boxes aren't good enough. The end effect is adults will have to upload their government papers to continue particip
1509.
by Buttons840 4y ago
We can't run the algorithm without the weights, be we can understand it -- or rather, we can understand it as much as it's possible for anyone to understand it. Having the weights will not add understanding. Having the loss func
1510.
by Buttons840 4y ago
I never would have guessed computers writing code would come before generating convincing voices.
1511.
by Buttons840 4y ago
Ah we're starting to bootstrap. For decades in reinforcement learning we've had Q learning, which promises to solve any optimization problem if only we can build a powerful enough function approximator. It can even learn off-p
1512.
by Buttons840 4y ago
After recently doing some JavaScript work and not knowing how to debug it, any suggestions for a JavaScript debugger? Part of the problem is it's already difficult to get a modern JavaScript stack setup: TypeScript transpiles to JavaSc
1513.
by Buttons840 4y ago
Being able to learn within context, without updating weights is amazing. Imagine how much more efficient and/or powerful it could be if we found a way to update the weights in real time.
1514.
by Buttons840 4y ago
> most people are lazy Most people collectively hold less than 1 or 2 percent of the stocks. People and funds who have 100 billion invested won't be lazy.
1515.
by Buttons840 4y ago
Just to let you know, scam alarms are going off in my head. You comment offering what seems to be an alternative view/opinion, and promoting your own work and research papers. When asked for more details the only thing you do is ask pe
1516.
by Buttons840 4y ago
I've been reading this paper with pseudocode for various transformers and finding it helfpul: https://arxiv.org/abs/2207.09238 "This document aims to be a self-contained, mathematically precise overview of tr
1517.
by Buttons840 4y ago
Headlines like this hint towards AI improving itself. Prompting itself in this case. But as we see in reinforcement learning, algorithms that act and improve themselves are not new. The interesting thing will be weather or not they eventual
1518.
by Buttons840 4y ago
I see, so [1, 2, 3, 4, 5, 6, 7, 8] is more similar to [5, 3, 1, 7, 8, 2, 4, 6] than with a linear layer? That's what you mean by permutation equivariant? I understand each context input is embedded with its position, but I suppose the
1519.
by Buttons840 4y ago
Do these scale as well as transformers? My understanding is that classic RNNs don't scale well, and that is one reason why transformers became popular. As a pleb who doesn't even own a data center, I've been hoping that a sup
1520.
by Buttons840 4y ago
If I want to do sequence modelling, let's say, predict the 9th element of the following sequence: [1, 2, 3, 4, 5, 6, 7, 8, 9] -- that is, I know the 8 most recent tokens as my "context", and I want to predict the next, 9 in t
1521.
by Buttons840 4y ago
> getting shit done A team that gets shit done will later look back and see that what they've done is shit.
1522.
by Buttons840 4y ago
> I understand the kind of person they’re trying to avoid hiring by accident I think I might be that kind, firmly in the ivory tower. I always feel frustrated and isolated because people just don't engage and acknowledge my concerns
1523.
by Buttons840 4y ago
It's all about cycle length. CO2 from shallow ground cycles into the air, then it quickly cycles back into the ground. The forest burns, the forest regrows. CO2 leaves, CO2 returns. Fossil fuels are also a cycle. We bring up the fossil
1524.
by Buttons840 4y ago
GTP man pages would be awesome. This is pretty close to that for GDB. Although I suspect this is a limited view of the possibilities akin to thinking "the internet sounds neat, I can use it to find new pen pals". It also makes me
1525.
by Buttons840 4y ago
Headline mentions "heart", but the word does not appear in the article. How will a vaccine fix plaque buildup? Or fix plaque breaking and causing strokes?
1526.
by Buttons840 4y ago
Dumbo is a Disney character, every bit as much of a character as Micky Mouse or others, just less popular. Of course, by the time Disney notices you you'll already have succeeded, but you may be forced to change brand. You can do it no
1527.
by Buttons840 4y ago
Can I ignore half the story and just focus on those fucking check-in tablets? The first time a clinic asked me to do a pre-checking using a web page, I did it, filled out about 75 questions in some mediocre web app and then I was ready when
1528.
by Buttons840 4y ago
The average yearly cost of raising a child is $15,000 (rounded down), the average yearly cost of owning an iPhone is $3,000 (rounded way up). People exist in that gap who can afford an iPhone, but cannot afford children -- even if they
1529.
by Buttons840 4y ago
> If the argument is that they are destitute To be clear, that is not the argument. The argument is over your claim that if someone can afford an iPhone then they can afford to have children.
1530.
by Buttons840 4y ago
The Little Learner[0]: Learn the basics using scheme. This is fundamental; no frameworks, just Scheme and functions. From here you can go anywhere. Then watch Neural Networks: Zero to Hero[1] to learn how autograd works, learn a few advance

More ›