Science!! Fucking magnets, how do they work?

iannis

iannis

Musty Nester
31,351
17,656
Tuco said:
Wait what, deep learning is dead?
Click to expand...
not quantum deep learning though!

I think he's just making an observation about buzzwords.

Deep learning has already evolved though, hasn't it? I don't think they're doing translators with it. And tbh, language translation is one of the most impressive ai feats I'm aware of. Even if it's mostly a fancier babelfish.
 
Tuco

Tuco

I got Tuco'd!
<Gold Donor>
45,398
73,467
iannis said:
not quantum deep learning though!

I think he's just making an observation about buzzwords.

Deep learning has already evolved though, hasn't it? I don't think they're doing translators with it. And tbh, language translation is one of the most impressive ai feats I'm aware of. Even if it's mostly a fancier babelfish.
Click to expand...
I don't know what you mean by "already evolved" because for deep learning, sky's the limit.
ap110520018042_custom-8fbb1ec5b25461f081f09a9464d46e66962126a4.jpg


It's just pretty hard to get there and increasing the input domain makes the difficulty/cost grow exponentially.
 
iannis

iannis

Musty Nester
31,351
17,656
Evolved past the techniques they were calling deep learning five years ago.

I'm not very familiar at all, but I do know the phrase deep learning gets used a lot less in the popular ai for dummies articles.
 
wormie

Asshat wormie

2023 Asshat Award Winner
<Gold Donor>
16,820
30,963
Tuco said:
We probably agree then. I think the massive glut of papers like that is just because you don't need to do shit to get that 0.002%, so you can coattail ride all day without producing anything novel or worth reading.

Most real-world computer science before deep-learning is more closed-form where once you find a better solution you've got something novel, so there's less room to needledick your way into paper acceptance by tweaking something and curating your data harder.
Click to expand...
I think a lot of it has to do with the industry demand and the money that goes with it. Any paper that you can get accepted into a conference is another step closer to the big $$$. Also, I honestly think the structure of peer review in CS field is hurting it. While using conferences instead of journals has facilitated a more open science approach to things, its also resulted in all research being reviewed in a speedy manner due to the massive amounts of papers and the small number of reviewers willing to work for the increasing number of conferences. This leads to a shortage of time and man power and has resulted in more weight being given to metrics and reported results with the actual progress of the scientific thought in the research becoming a distant secondary. It may also be the fact that CS went from being a mathish research field to being a scientific one and the people doing review are not equipped with the experience of doing experimental science. Whatever it is, the current environment in this field cant last and will soonish start producing crap that people will turn on and the result will be another DL winter.
 
Tuco

Tuco

I got Tuco'd!
<Gold Donor>
45,398
73,467
iannis said:
Evolved past the techniques they were calling deep learning five years ago.

I'm not very familiar at all, but I do know the phrase deep learning gets used a lot less in the popular ai for dummies articles.
Click to expand...
I don't know about those articles, but "deep learning" is a pretty broad term that will encapsulate pretty much every kind of approach that uses lots of data to produce a custom-built processing system from some kind of very basic mathematical principles like convolutions.
 
  • 1Like
Reactions: 1 user
Tuco

Tuco

I got Tuco'd!
<Gold Donor>
45,398
73,467
wormie said:
I think a lot of it has to do with the industry demand and the money that goes with it. Any paper that you can get accepted into a conference is another step closer to the big $$$. Also, I honestly think the structure of peer review in CS field is hurting it. While using conferences instead of journals has facilitated a more open science approach to things, its also resulted in all research being reviewed in a speedy manner due to the massive amounts of papers and the small number of reviewers willing to work for the increasing number of conferences. This leads to a shortage of time and man power and has resulted in more weight being given to metrics and reported results with the actual progress of the scientific thought in the research becoming a distant secondary. It may also be the fact that CS went from being a mathish research field to being a scientific one and the people doing review are not equipped with the experience of doing experimental science. Whatever it is, the current environment in this field cant last and will soonish start producing crap that people will turn on and the result will be another DL winter.
Click to expand...
Sounds more like growing pains than a deflation of a field though.

I think I take a longer stance of, "Is shit going to be much better in 5 or 10 years?" and my answer is "fuck yeah it will". It might do so on the blood and sweat of the poor sons of bitches that have to read hundreds of regurgitated papers trying to find the novel needle in the haystack, but fuck those guys anyway am I right?

Speaking personally as a guy who can skim the proven tech that might've been made ~2-3 years ago, and use it to solve real-world problems, deep learning is great and hype is high.
 
iannis

iannis

Musty Nester
31,351
17,656
Might be why they quit using it. Sounds like magic to the ignorant, sounds ignorant to the educated.
 
  • 1Like
Reactions: 1 user
wormie

Asshat wormie

2023 Asshat Award Winner
<Gold Donor>
16,820
30,963
Tuco said:
Sounds more like growing pains than a deflation of a field though.

I think I take a longer stance of, "Is shit going to be much better in 5 or 10 years?" and my answer is "fuck yeah it will". It might do so on the blood and sweat of the poor sons of bitches that have to read hundreds of regurgitated papers trying to find the novel needle in the haystack, but fuck those guys anyway am I right?

Speaking personally as a guy who can skim the proven tech that might've been made ~2-3 years ago, and use it to solve real-world problems, deep learning is great and hype is high.
Click to expand...
Growing pains sounds very likely. Once its grown and becomes established, it stops being novel and stops being hyped.
 
  • 2Like
Reactions: 1 users
pharmakos

pharmakos

soʞɐɯɹɐɥd
<Bronze Donator>
16,306
-2,239
wormie said:
It’s not an individual blog post you stupid fuck. It’s about hyping up unproven and untested shit in order to get clicks. What happens when this piece of science is debunked? How will that affect public perception of use of tax funds? You are right that science needs to be presented to a wide audience in order to continue its support. You are a stupid mother fucker for thinking that hyping up unknown science is in a useful way to go about it.
Click to expand...

You think this Ars post is potentially going to derail science and cause some drastic loss of funds to science in general, that's your counterargument now?

You are devolving into being so much more retarded than you initially called me out to be lol
 
wormie

Asshat wormie

2023 Asshat Award Winner
<Gold Donor>
16,820
30,963
pharmakos said:
You think this Ars post is potentially going to derail science and cause some drastic loss of funds to science in general, that's your counterargument now?

You are devolving into being so much more retarded than you initially called me out to be lol
Click to expand...
Here is my counterargument:

:emoji_fork_knife_plate::emoji_poop::emoji_coffin::emoji_gay_pride_flag:
 
  • 1Worf
Reactions: 1 user
Cybsled

Cybsled

Avatar of War Slayer
16,415
12,040
Next time someone complains about how US should be metric

www.mentalfloss.com

The Origins of 7 Common Units of Measurement

Ever wonder why there are 5280 feet in a mile?
www.mentalfloss.com www.mentalfloss.com

Imperial units had basis in the ROMAN EMPIRE, which means the US is the true successor to the greatest empire in history lol
 
  • 1Solidarity
  • 1Dislike
  • 1Like
Reactions: 2 users
Borzak

Borzak

Bronze Baron of the Realm
24,547
31,830
Every so often I still run across someones timber cruise and they have it listed by chains. 80 chains in a mile. My brain is full of useless shit.
 
pharmakos

pharmakos

soʞɐɯɹɐɥd
<Bronze Donator>
16,306
-2,239
imperial volume units make sense to me at least, just powers of 2 for the most part.
 
P

Pescador

Trakanon Raider
234
239
Omi43221 said:
Scientists admitting their work isn't reproducible. Hopefully this starts a trend.
www.forbes.com

Nobel Prize Winner Frances Arnold Retracts Paper, Here Is The Reaction

Arnold's move garnered praise on Twitter and showed how scientific research needs to change.
www.forbes.com www.forbes.com
Click to expand...
Buried a little deeper in the article is the real issue: scientists are rewarded for getting good results published, so there is no incentive to reproduce their work, especially if the paper is kind of a "one-off" set of studies.

Most scientists are short on time and short on money, and virtually all of their career success is linked to high-impact publications and successful grant applications. Reproducing work from successful studies won't really help you get either, so it's far more rewarding to move on to the next paper, especially in a competitive field where many groups are "racing" to the next milestone.

Some prominent researchers are lucky to reach a point in their careers where they can afford to slow down and check things, particularly when they are a big name with tons of finding and an army of grad students / post docs. But what about that young investigator who has spent 10-15 years working towards tenure? Retracting a Science article could literally sink their entire career, so not only is there no incentive to reproduce that work, there is actually every motivation to file it away and move on.

A lot of the reproducibility issues come to light when technologies move past the research stage and begin development. But for most academics, that is never even on their radar for their entire career.
 
  • 4Like
Reactions: 3 users
reavor

reavor

I'm With HER ♀
<Bronze Donator>
4,762
16,496
Pescador said:
Buried a little deeper in the article is the real issue: scientists are rewarded for getting good results published, so there is no incentive to reproduce their work, especially if the paper is kind of a "one-off" set of studies.

Most scientists are short on time and short on money, and virtually all of their career success is linked to high-impact publications and successful grant applications. Reproducing work from successful studies won't really help you get either, so it's far more rewarding to move on to the next paper, especially in a competitive field where many groups are "racing" to the next milestone.

Some prominent researchers are lucky to reach a point in their careers where they can afford to slow down and check things, particularly when they are a big name with tons of finding and an army of grad students / post docs. But what about that young investigator who has spent 10-15 years working towards tenure? Retracting a Science article could literally sink their entire career, so not only is there no incentive to reproduce that work, there is actually every motivation to file it away and move on.

A lot of the reproducibility issues come to light when technologies move past the research stage and begin development. But for most academics, that is never even on their radar for their entire career.
Click to expand...

Yep, the publish or perish mantra, and the disdain for negative results, is one of the main issues behind this. Also that reviewing papers is extracurricular work, you don't get paid, and you have to do it a lot before you can get into a nice role on an editor board for a journal or similar, which is why you can't spend too much time reviewing one single study. Another issue is that experiments are getting exponentially more intricate, complex, time-consuming and particularly expensive to perform.
 
Cybsled

Cybsled

Avatar of War Slayer
16,415
12,040
iannis said:
wormie wormie

QUANTUM DICK PILLS, BROTHER.

QUANTUM. DICK. PILLS.
Click to expand...

Wouldn’t a quantum dick pill mean it exists in 2 states simultaneously?

“this isn’t working!”

“sure it is. Your eyeballs just can not comprehend the multiple quantum states! That will be $500 no refunds”
 
  • 2Worf
Reactions: 1 users
You must log in or register to reply here.