kmmlng

7 ms·

31.
by kmmlng 2y ago
That seems to be in support of the parent's point. So breaking up Google would have resulted in worse outcomes for consumers 10 years ago, but it would be even worse today? Then it wouldn't have made sense to break them up 10 year
32.
by kmmlng 2y ago
I would disagree with the second point. Music theory has definitely changed how I make music. The initial idea is typically still conceived intuitively, but I definitely use music theory to develop it into a whole piece. Before knowing musi
33.
by kmmlng 2y ago
> Yes, but training is the most expensive part of ML, for example GPT-3 is estimated to cost something like 1-4 million USD. That entirely depends on how many inferences the model will perform during its lifecycle. You can find different
34.
by kmmlng 2y ago
It's very variable. 3 years is going to be the shortest duration, but it can be much longer.
35.
by kmmlng 2y ago
Isn't this similar to what Microsoft did with their Phi models?
36.
by kmmlng 2y ago
In academic circles, the abstract is typically a summary, not a problem statement. The rest of the sections are usually read out of order, so it's actually not that different.
37.
by kmmlng 2y ago
I feel like on other laptops everyone is optimizing for specs/price and specs don't include quality of life features like a good trackpad. Sure, it would be nice to have a good trackpad, but chances are your customers are going to
38.
by kmmlng 2y ago
I think any parent who thinks of their child as "worthless" has got to be at least partially at fault for whatever pathologies follow from that kind of attitude.
39.
by kmmlng 2y ago
Of course you can find stuff that's high quality, but that is rare. Rather than looking at outliers, it might be more sensible to look at the average and the reality is that the average graffiti does not have any artistic value.
40.
by kmmlng 2y ago
Eh, it really doesn't matter much in practice. Additionally, there are many other activation functions without this issue.
41.
by kmmlng 2y ago
How did you manage to secure a job where you get paid for life-long learning?
42.
by kmmlng 2y ago
I'm not sure how it is for the companies, but as universities involved in EU projects, the amount of bureaucracy is just overwhelming and there are so many nonsensical requirements, that you don't get to do any interesting work. O
43.
by kmmlng 3y ago
In my experience, this is only a problem when you are using Excel. It's ridiculous how bad Excel is at handling CSVs, I really cannot comprehend it. If you use LibreOffice all your problems magically disappear.
44.
by kmmlng 3y ago
The interesting thing is that they are still pushing the limits in new ways. Will there be someone else in 10 years that can do what they are doing now? Sure, but SpaceX will then also be ahead of current SpaceX by 10 years. I don't kn
45.
by kmmlng 3y ago
While I agree that moving all human interaction to the internet is likely to be a net negative, there is only so much you can do as an individual. It takes two to tango, you need a sufficient number of other people with the same mindset to
46.
by kmmlng 3y ago
It causes lots of problems because many of our systems have been set up in times of growth and won't function without it. Take for example "pay-as-you-go" pension systems in much of Europe. Here, you don't fund your own
47.
by kmmlng 3y ago
I remember this transition and funnily enough, the shorter battery life actually made for better user experience. Before, your battery would run out at some point . It was always kind of unexpected and then you couldn't use your phone
48.
by kmmlng 3y ago
As conceptualized by Seligman, there is also a component of optimism called "explanatory style". Basically, if you experience some kind of adverse event, how do you explain to yourself what has happened? Let's say you are rej
49.
by kmmlng 3y ago
What do you mean, the development of the "Perception"? Do you mean the Perceptron? In that case, Backprop was invented way later than the Perceptron (see https://people.idsia.ch/~juergen/who-invented-backpropa
50.
by kmmlng 3y ago
I find "Work" to be a bit of an odd inclusion here. Presumably most people have a job and necessarily complete many work tasks each day?
51.
by kmmlng 3y ago
> This is all true in a neutral net, but Transformers aren't Neural Nets in the traditional sense. I was under that impression originally, but there's not a back propagation or Hebbian learning here, which were the key bits of
52.
by kmmlng 3y ago
Oh no, the evidence is clear when it comes to smoking. What I was trying to say: Smoking is a nice reference point to orient yourself around, because you can fairly confidently conclude that it's very harmful even with imperfect method
53.
by kmmlng 3y ago
The fact that you can pick up on negative effects through observational studies if they are large enough makes me especially suspicious about results that differ from study to study. Perhaps there is an effect there, but if we can't re
54.
by kmmlng 3y ago
While this is true, in some sense it is also a feature and not a bug. You don't always want the whole world seeing what you're writing. While Discord doesn't offer complete privacy (unless you restrict access), it provides go
55.
by kmmlng 3y ago
This stuff is all about risk. If two people add their personal experiences, we have two data points. What are we supposed to do with that? It's not enough to quantify risk. You can do everything wrong and be completely fine or you can
56.
by kmmlng 3y ago
To make a somewhat exaggerated comparison: horses still work fine today, but it really doesn't make sense for people to be using them anymore outside of niche use cases or as a hobby. Sure, macs from 20 years ago still work "fine&
57.
by kmmlng 3y ago
Absolutely, the organization as a whole is succesful, but the exact same people in a different organization might not quite be as succesful. There is something in the DNA of the organization that is just working and that is an effect of how
58.
by kmmlng 3y ago
Yes, the label "peer review" is meaningless. There are good peer review processes at top journals and conferences and getting through those is a meaningful accomplishment. Then there are also worthless peer review processes at low
59.
by kmmlng 3y ago
This is true, but as in many areas, power laws apply. The vast majority of research has very low impact and mostly serves fulfilling publication quotas. There is likely plenty of scientific fraud happening in lower tier publications as well
60.
by kmmlng 3y ago
I would consider a well-defined mean and standard deviation an assumption. The distribution of maximum entropy is determined by the constraints. Those constraints have to be assumed. If you constrain your problem to only have nonzero probab

More ›