Go Back   Grim Dawn Forums > Off Topic > Rants

Reply
 
Thread Tools Display Modes
  #21  
Old 03-15-2018, 10:06 AM
Ptirodaktill's Avatar
Ptirodaktill Ptirodaktill is offline
Herald
 
Join Date: Dec 2015
Posts: 1,370
Default

Quote:
Originally Posted by Superfluff View Post
I reedited the prices.

So you would chose that I3 over a ryzen 5? That has pretty much the same price where I live
It depend on how fast Intel could deploy cheap MOBO, cause right now all i see from intel are z370 monsters for 220$+ >_> As for Ryzen you can buy AM4 mobo for 60$ and it should work fine.
__________________
#TeamDoubleDemo
#HowDoIRei:frog:
Reply With Quote
  #22  
Old 03-15-2018, 10:12 AM
Superfluff's Avatar
Superfluff Superfluff is online now
Praetorian
 
Join Date: Jun 2016
Posts: 4,419
Default

Quote:
Originally Posted by Ptirodaktill View Post
It depend on how fast Intel could deploy cheap MOBO, cause right now all i see from intel are z370 monsters for 220$+ >_> As for Ryzen you can buy AM4 mobo for 60$ and it should work fine.
Forgot about the mobo part. I'd like to keep that in the 120$ area max.

This model "Asus PRIME B350-PLUS, Socket AM4" seems to have good reviews

A cheap one at half the price would be "MSI B350M PRO-VD PLUS, socket AM4". Also seems decent
Reply With Quote
  #23  
Old 03-15-2018, 10:23 AM
Ptirodaktill's Avatar
Ptirodaktill Ptirodaktill is offline
Herald
 
Join Date: Dec 2015
Posts: 1,370
Default

Quote:
Originally Posted by Superfluff View Post
Forgot about the mobo part. I'd like to keep that in the 120$ area max.

This model "Asus PRIME B350-PLUS, Socket AM4" seems to have good reviews

A cheap one at half the price would be "MSI B350M PRO-VD PLUS, socket AM4". Also seems decent
MSI B350M PRO-VD PLUS is good!
__________________
#TeamDoubleDemo
#HowDoIRei:frog:
Reply With Quote
  #24  
Old 03-15-2018, 10:46 AM
Superfluff's Avatar
Superfluff Superfluff is online now
Praetorian
 
Join Date: Jun 2016
Posts: 4,419
Default

Quote:
Originally Posted by Ptirodaktill View Post
MSI B350M PRO-VD PLUS is good!
Thanks for the tips!
Reply With Quote
  #25  
Old 03-17-2018, 09:59 AM
durruti's Avatar
durruti durruti is offline
Advocate
 
Join Date: Mar 2012
Posts: 493
Default

Quote:
Originally Posted by Trudel View Post
The article makes it plain, that manipulation of AMDs stock price was the motive behind the release of report. Critics were cited. They reached out to a researcher who verified the claims.
and their title even says that it makes bad hacks worse, i.e. they are secondary exploits. i guess you're right, i was still in rage.

trail of bits credibility kinda hangs on cts-labs' though (turns out they were paid to verify the findings which makes sense but doesn't help cts-labs point, that they really couldn't have given amd more time - or any time at all...) the other supportive researcher had not seen any poc and was just tweeting bs at the time.

so some critical reports later the bs from cts-labs goes on and on.

Quote:
Originally Posted by Superfluff View Post
Thanks for the tips!
do avoid boards with asmedia-controllers, those started it all.

Quote:
Originally Posted by Sneaky parrot View Post
anyone who knows a tiny bit of modern nn knows how brittle the architecture really is.
lol, go play go vs alphazero. not sure brittle is the right word. i'd guess it is not so much about the architecture of neural nets but that a trained one is a whole different beast.

Last edited by durruti; 03-17-2018 at 10:40 AM.
Reply With Quote
  #26  
Old 03-17-2018, 02:30 PM
Sneaky parrot's Avatar
Sneaky parrot Sneaky parrot is offline
Supporter
 
Join Date: Jan 2018
Posts: 167
Default

Quote:
Originally Posted by durruti View Post
and their title even says that it makes bad hacks worse, i.e. they are secondary exploits. i guess you're right, i was still in rage.

trail of bits credibility kinda hangs on cts-labs' though (turns out they were paid to verify the findings which makes sense but doesn't help cts-labs point, that they really couldn't have given amd more time - or any time at all...) the other supportive researcher had not seen any poc and was just tweeting bs at the time.

so some critical reports later the bs from cts-labs goes on and on.


do avoid boards with asmedia-controllers, those started it all.


lol, go play go vs alphazero. not sure brittle is the right word. i'd guess it is not so much about the architecture of neural nets but that a trained one is a whole different beast.

Dude, u know that all of them are prone to adversarial attacks and there is not a good solution right now right?
Reply With Quote
  #27  
Old 03-17-2018, 04:46 PM
durruti's Avatar
durruti durruti is offline
Advocate
 
Join Date: Mar 2012
Posts: 493
Default

Quote:
Originally Posted by Sneaky parrot View Post
Dude, u know that all of them are prone to adversarial attacks and there is not a good solution right now right?
so it's either elons fear of ai going rogue or that they become ubiquitous while being easily tricked because their real world application was framed in shitty mathematical terms. i thought you wanted to dispel fears.
Reply With Quote
  #28  
Old 03-17-2018, 08:20 PM
Dioarchet Dioarchet is offline
Emissary
 
Join Date: Sep 2015
Posts: 943
Default

I think both of you guys are right about current NN's.

AlphaGo was a success partially because they didn't have to worry about adversarial training data. Who in the history of Go played a game for the purposes of fooling some AI?

There ought to be plenty of other applications for current NN's that don't have to worry about adversarial data or where that risk can be mitigated with a variety of techniques. Obviously for military applications and other competitive endeavors, the risk is much higher, perhaps unacceptably so.

I'm sure both of you have heard of Hinton's CapsNet by now, which promises to avoid much of the "brittleness" of current NN's although it's still not immune to adversarial attacks. Still, it goes to show that the latest state-of-the-art may hardly remain so for long, and that there may still be breakthroughs that get us closer to what Elon's envisioning.

And, aren't humans also vulnerable to "adversarial attacks?" When someone gets fooled by a phishing email or gets wowed by some magician on stage, that's not hugely conceptually different than some NN mis-categorizing some object. And yet, no one thinks humans aren't ever a "good solution" or that they're always "brittle."

Both humans and AI have weaknesses right now, but they also have their strengths. The key is to use the right "tool" for the job.

Last edited by Dioarchet; 03-17-2018 at 08:43 PM.
Reply With Quote
  #29  
Old 03-18-2018, 12:35 AM
Sneaky parrot's Avatar
Sneaky parrot Sneaky parrot is offline
Supporter
 
Join Date: Jan 2018
Posts: 167
Default

Quote:
Originally Posted by durruti View Post
so it's either elons fear of ai going rogue or that they become ubiquitous while being easily tricked because their real world application was framed in shitty mathematical terms. i thought you wanted to dispel fears.
Elon is more like Edison on this matter.......I remember Edison said something 'nice' about AC to eliminate competition. There is so much you can do with the current nn , but the margin for error is pretty small.
Reply With Quote
  #30  
Old 03-18-2018, 12:45 AM
Sneaky parrot's Avatar
Sneaky parrot Sneaky parrot is offline
Supporter
 
Join Date: Jan 2018
Posts: 167
Default

Quote:
Originally Posted by Dioarchet View Post
I think both of you guys are right about current NN's.

AlphaGo was a success partially because they didn't have to worry about adversarial training data. Who in the history of Go played a game for the purposes of fooling some AI?

There ought to be plenty of other applications for current NN's that don't have to worry about adversarial data or where that risk can be mitigated with a variety of techniques. Obviously for military applications and other competitive endeavors, the risk is much higher, perhaps unacceptably so.

I'm sure both of you have heard of Hinton's CapsNet by now, which promises to avoid much of the "brittleness" of current NN's although it's still not immune to adversarial attacks. Still, it goes to show that the latest state-of-the-art may hardly remain so for long, and that there may still be breakthroughs that get us closer to what Elon's envisioning.

And, aren't humans also vulnerable to "adversarial attacks?" When someone gets fooled by a phishing email or gets wowed by some magician on stage, that's not hugely conceptually different than some NN mis-categorizing some object. And yet, no one thinks humans aren't ever a "good solution" or that they're always "brittle."

Both humans and AI have weaknesses right now, but they also have their strengths. The key is to use the right "tool" for the job.
That's pretty much it.
Remember the time when John McCarthy boldly claimed that AI will dominate the world to get government fundings....
Yeah,I'm more curious how Elon grows his hair back....(XDXDXD)
Reply With Quote
Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


Grim Dawn ©2018 Crate Entertainment, LLC.
vBulletin® 3.8.4 ©2000-2014, Jelsoft Enterprises Ltd.