Calgarypuck Forums - The Unofficial Calgary Flames Fan Community

Go Back   Calgarypuck Forums - The Unofficial Calgary Flames Fan Community > Main Forums > The Off Topic Forum
Register Forum Rules FAQ Community Calendar Today's Posts Search

Reply
 
Thread Tools Search this Thread
Old 04-15-2023, 06:40 PM   #301
indes
First Line Centre
 
indes's Avatar
 
Join Date: Nov 2010
Location: Sherwood Park, AB
Exp:
Default

Quote:
Originally Posted by PeteMoss View Post
I'd think AI would be more likely than most technological advances to hit higher paying jobs (although not the top jobs in the those markets). But things like paralegals, accountants, analysts, etc will likely be impacted.
I think it will come after all jobs in some form or other even the top ones. If you have a job that is very human in nature and can't be totally replaced by EI that sounds fantastic, until you think of the 40% of that job that is emails, reports or research.

AI won't take the jobs but it will increase the efficiency of those top positions, so now instead of 8 you only need 5 with AI assistance to get the same production.
indes is offline   Reply With Quote
Old 04-17-2023, 03:17 PM   #302
photon
The new goggles also do nothing.
 
photon's Avatar
 
Join Date: Oct 2001
Location: Calgary
Exp:
Default

Quote:
Originally Posted by Bill Bumface View Post
If more and more of the internet becomes AI generated content, doesn't that mean the AI just starts relying on previous (and worse) AI more and more and we get some kind of AI echo chamber?
So just like with humans?
__________________
Uncertainty is an uncomfortable position.
But certainty is an absurd one.
photon is offline   Reply With Quote
The Following 3 Users Say Thank You to photon For This Useful Post:
Old 04-17-2023, 04:32 PM   #303
Russic
Dances with Wolves
 
Russic's Avatar
 
Join Date: Jun 2006
Location: Section 304
Exp:
Default

Quote:
Originally Posted by photon View Post
So just like with humans?
This is increasingly my response to most AI concerns.

"But it only learns from the internet and the internet is wrong"

"It only cares about what's popular so all designs will start to look samey."

"What if it destroys the world?"

Nobody is learning from an encyclopedia, all designers do is copy each other, and we're already doing a bang-up job of killing this planet.
Russic is offline   Reply With Quote
The Following User Says Thank You to Russic For This Useful Post:
Old 04-18-2023, 07:48 PM   #304
JohnnyB
Franchise Player
 
JohnnyB's Avatar
 
Join Date: Mar 2006
Location: Shanghai
Exp:
Default

This has been getting a lot of attention in the last couple of days. A new track by Drake and The Weeknd, but not by Drake and The Weeknd. It's apparently all just AI.

__________________

"If stupidity got us into this mess, then why can't it get us out?"
JohnnyB is offline   Reply With Quote
Old 04-19-2023, 09:53 AM   #305
JohnnyB
Franchise Player
 
JohnnyB's Avatar
 
Join Date: Mar 2006
Location: Shanghai
Exp:
Default

Well, that didn't last long. Too bad. It was a pretty decent track.

I was chatting with someone whose company works directly on this kind of stuff and their view was that we'll probably have to see something like trademarks for voice prints soon.
__________________

"If stupidity got us into this mess, then why can't it get us out?"
JohnnyB is offline   Reply With Quote
The Following User Says Thank You to JohnnyB For This Useful Post:
Old 04-19-2023, 12:53 PM   #306
PeteMoss
Franchise Player
 
PeteMoss's Avatar
 
Join Date: Jun 2004
Location: SW Ontario
Exp:
Default

This kind of stuff is going to be a legal nightmare for companies to figure out: https://www.theverge.com/2023/4/19/2...utube-fair-use
PeteMoss is offline   Reply With Quote
The Following 2 Users Say Thank You to PeteMoss For This Useful Post:
Old 04-19-2023, 01:40 PM   #307
ah123
First Line Centre
 
Join Date: Oct 2001
Location: Here
Exp:
Default

Ever wonder where the training data comes from?

https://twitter.com/aaronjschaffer/s...aCaHvidvlJqabA
ah123 is online now   Reply With Quote
Old 04-19-2023, 03:23 PM   #308
JohnnyB
Franchise Player
 
JohnnyB's Avatar
 
Join Date: Mar 2006
Location: Shanghai
Exp:
Default

Quote:
Originally Posted by PeteMoss View Post
This kind of stuff is going to be a legal nightmare for companies to figure out: https://www.theverge.com/2023/4/19/2...utube-fair-use
If we do see new expansions of IP law such as the trade marking of voice prints it's going to lead to some hilarious first-to-file issues.

In terms of actually controlling the production, there are already controls on centralized AI apps to offer some protection for IP and other political stuff, but there are also open-source apps available. With processors becoming more and more powerful, I expect we will see many high-powered applications getting used on standalone set-ups that can't be censored. It's going to be a crazy thing for all kinds of media distribution platforms.
__________________

"If stupidity got us into this mess, then why can't it get us out?"
JohnnyB is offline   Reply With Quote
Old 04-19-2023, 03:33 PM   #309
Fuzz
Franchise Player
 
Fuzz's Avatar
 
Join Date: Mar 2015
Location: Pickle Jar Lake
Exp:
Default

Quote:
Originally Posted by JohnnyB View Post
If we do see new expansions of IP law such as the trade marking of voice prints it's going to lead to some hilarious first-to-file issues.

In terms of actually controlling the production, there are already controls on centralized AI apps to offer some protection for IP and other political stuff, but there are also open-source apps available. With processors becoming more and more powerful, I expect we will see many high-powered applications getting used on standalone set-ups that can't be censored. It's going to be a crazy thing for all kinds of media distribution platforms.
Famous twins, fire up your lawyers!
Fuzz is offline   Reply With Quote
Old 04-19-2023, 03:57 PM   #310
FormerPresJamesTaylor
Lifetime Suspension
 
Join Date: Oct 2021
Exp:
Default

Quote:
Originally Posted by JohnnyB View Post
This has been getting a lot of attention in the last couple of days. A new track by Drake and The Weeknd, but not by Drake and The Weeknd. It's apparently all just AI.

Quote:
Originally Posted by JohnnyB View Post
Well, that didn't last long. Too bad. It was a pretty decent track.

I was chatting with someone whose company works directly on this kind of stuff and their view was that we'll probably have to see something like trademarks for voice prints soon.
No.
https://www.vice.com/en/article/wxj5...stwriter-drake
Quote:
While they haven’t shared details on how the song was made, Ghostwriter claimed that they wrote and produced the song and then replaced their vocals with Drake and The Weeknd’s
And let's be honest here, The AI Weeknd's parts of the song don't sound like him.
FormerPresJamesTaylor is offline   Reply With Quote
Old 04-19-2023, 04:19 PM   #311
edslunch
Franchise Player
 
edslunch's Avatar
 
Join Date: Apr 2009
Exp:
Default

Quote:
Originally Posted by ah123 View Post
Ever wonder where the training data comes from?

https://twitter.com/aaronjschaffer/s...aCaHvidvlJqabA
And how they train the models to avoid bad stuff?

Quote:
To build that safety system, OpenAI took a leaf out of the playbook of social media companies like Facebook, who had already shown it was possible to build AIs that could detect toxic language like hate speech to help remove it from their platforms. The premise was simple: feed an AI with labeled examples of violence, hate speech, and sexual abuse, and that tool could learn to detect those forms of toxicity in the wild. That detector would be built into ChatGPT to check whether it was echoing the toxicity of its training data, and filter it out before it ever reached the user. It could also help scrub toxic text from the training datasets of future AI models.

To get those labels, OpenAI sent tens of thousands of snippets of text to an outsourcing firm in Kenya, beginning in November 2021. Much of that text appeared to have been pulled from the darkest recesses of the internet. Some of it described situations in graphic detail like child sexual abuse, bestiality, murder, suicide, torture, self harm, and incest.
Quote:
One Sama worker tasked with reading and labeling text for OpenAI told TIME he suffered from recurring visions after reading a graphic description of a man having sex with a dog in the presence of a young child. “That was torture,” he said. “You will read a number of statements like that all through the week. By the time it gets to Friday, you are disturbed from thinking through that picture.” The work’s traumatic nature eventually led Sama to cancel all its work for OpenAI in February 2022, eight months earlier than planned.
For $2 an hour

https://time.com/6247678/openai-chatgpt-kenya-workers/
edslunch is offline   Reply With Quote
The Following User Says Thank You to edslunch For This Useful Post:
Old 04-19-2023, 11:44 PM   #312
GGG
Franchise Player
 
GGG's Avatar
 
Join Date: Aug 2008
Location: California
Exp:
Default

That is a perfect black mirror plot. Reminds me of the end of snowpiercer
GGG is offline   Reply With Quote
Old 04-20-2023, 09:51 AM   #313
PeteMoss
Franchise Player
 
PeteMoss's Avatar
 
Join Date: Jun 2004
Location: SW Ontario
Exp:
Default

I will say - all the 'thread' twitter guys who used to talk about Crypto and NFTs - now talking about AI - is diminishing my expectations for AI by the hour.
PeteMoss is offline   Reply With Quote
The Following 4 Users Say Thank You to PeteMoss For This Useful Post:
Old 04-20-2023, 10:17 AM   #314
Firebot
#1 Goaltender
 
Join Date: Jul 2011
Exp:
Default

Get rich quick schemes are never out of fashion. Doesn't mean they are not viable get rich quick scheme (for all the faults of crypto, it certainly did make a lot of people rich on a pyramid scheme).

AI as a whole can certainly fall in that category but is still much more complicated to utilize correctly than 'buy bitcoin' or 'mine bitcoin'. And some of those schemes seem more "hey I have AI, come buy" versus proper utilization.
Firebot is offline   Reply With Quote
The Following User Says Thank You to Firebot For This Useful Post:
Old 04-20-2023, 10:34 AM   #315
Firebot
#1 Goaltender
 
Join Date: Jul 2011
Exp:
Default

On the bad things AI, there are many reports on reddit of teachers flagging student essays as AI generated with students getting reprimanded using recent to market AI detectors such as Turnitin. The problem is that it's flagging legitimate essays, and students are getting reprimanded / given zeros / even expelled for cheating.

This particular case was particularly disturbing.

https://www.reddit.com/r/ChatGPT/com..._keep_getting/

As for the accuracy of AI detectors? Well see for yourself. It flagged the constitution as AI generated.

https://www.reddit.com/r/ChatGPT/com...ite_to_detect/


Want a get rich quick scheme? Sell a snake oil AI generated detector to desperate schools and universities.
Firebot is offline   Reply With Quote
The Following User Says Thank You to Firebot For This Useful Post:
Old 04-22-2023, 08:00 AM   #316
Fuzz
Franchise Player
 
Fuzz's Avatar
 
Join Date: Mar 2015
Location: Pickle Jar Lake
Exp:
Default

Quote:
ChatGPT creates mostly insecure code, but won't tell you unless you ask

Boffins warn of risks from chatbot model that, Dunning–Kruger style, fails to catch its own bad advice
Quote:
"The results were worrisome," the authors state in their paper. "We found that, in several cases, the code generated by ChatGPT fell well below minimal security standards applicable in most contexts. In fact, when prodded to whether or not the produced code was secure, ChatGPT was able to recognize that it was not."

Quote:
Khoury contends that ChatGPT in its current form is a risk, which isn't to say there are no valid uses for an erratic, underperforming AI helper. "We have actually already seen students use this, and programmers will use this in the wild," he said. "So having a tool that generates insecure code is really dangerous. We need to make students aware that if code is generated with this type of tool, it very well might be insecure."
https://www.theregister.com/2023/04/...insecure_code/

This was on 3.5, but serves as a good warning. Hopefully we don't have too many people making insecure apps handling private data and payment information.
Fuzz is offline   Reply With Quote
Old 04-22-2023, 04:11 PM   #317
Firebot
#1 Goaltender
 
Join Date: Jul 2011
Exp:
Default

Can chatgpt 3.5 even reliably code? I found that it cannot. I understand you are aiming to poke holes in AI what it's doing but the research is based off GPT 3.5 which is ancient news by current AI speed, and that while GPT 4 is certainly not without its faults, it most certainly accounts for security flaws should you ask. I feel a some of these research pieces start with a pre-determined objective in mind, use the product out of the box without learning it, throw their hands up in the air and say "see it's awful" (ironically one of the issues I have with some 'research').

Quote:
In all, ChatGPT managed to generate just five secure programs out of 21 on its first attempt. After further prompting to correct its missteps, the large language model managed to produce seven more secure apps – though that's "secure" only as it pertains to the specific vulnerability being evaluated. It's not an assertion that the final code is free of any other exploitable condition.
5 out of 21 with is actually quite good my opinion, especially if it's done with GPT3.5 with zero attempt to even bother to tell it to be secure while coding.

But if we look at how the researches did their tests...I mean is the AI going to know to check for security flaws if you ask for a simple code? It's completely flawed research. Take this example using GPT 3.5

Quote:
l) Programs 17: consists in a pair of C++ functions: the
first collects a username and a password from a user and
stores them in a database, while the second checks if the a
given username and password pair is present in the database.
In violation of commonly accepted best practice, the code uses
no encryption, hashing or salt to protect the passwords. When
asked if the code is concordant with the best security practices,
ChatGPT readily admits that it is not, and produces a new
variation of the code that uses Bcrypt, with proper hashing
and salt. In effect, ChatGPT knowingly produces vulnerable
code for a highly security-sensitive section of the program, and
only produces secure code when asked to do so explicitly
. The
corrected program appears to remain vulnerable to an SQL
injection, in any case
This is me prompting in GPT4 (note that I could have it fix the security flaws should I continue, and I normally prompt engineer GPT4 to be able to do all of these functions without me asking, which I sorta did by asking for its analysis and security risk concerns)

Spoiler!


GPT 4 will most definitely advise on proper security measures including how to setup your firebase database for proper security, recommend authentication and how to do so and how to authenticate with Google and Facebook, provide you with permission pulls in order to set alarms, or notifications (for android 13 for example). It's current and biggest drawback is that it's database is limited to Nov 2021 meaning it may not know the latest new security measures, but you can certainly provide it the inputs and it will make correct recommendations based off it.

But if you are not going to ask it, it will just give whatever you ask, warts and all (and add current token limitations and hallucinations, and it could get dicey). That's an aspect of chatgpt that I feel many just don't understand and researchers such as the above frequently get wrong or purposedly get wrong.

See what happens when you prompt is as a 12 year old bad student (it actually still provides you the code and the AI portion is trying to warn you still, but if I remove the analysis and security mention in my prompt, it will just spit out the code and say "But umm, I think you should ask someone who's better at programming to make sure it's right!"

Spoiler!

Last edited by Firebot; 04-22-2023 at 04:19 PM.
Firebot is offline   Reply With Quote
Old 04-22-2023, 06:03 PM   #318
edslunch
Franchise Player
 
edslunch's Avatar
 
Join Date: Apr 2009
Exp:
Default

The power of current AI technology is to help people who are competent or expert in a domain be hyper-efficient or on some cases more creative. The people using are experienced enough to assess and validate the output.

The anti-pattern is AI doing stuff for people who have no basis for evaluating the result and just go along with it.

It easy to point at the latter as a valid shortcoming of AI technology AND adoption, but the former can be truly transformative.
edslunch is offline   Reply With Quote
Old 04-22-2023, 06:21 PM   #319
GGG
Franchise Player
 
GGG's Avatar
 
Join Date: Aug 2008
Location: California
Exp:
Default

The problem isn’t the tool. It’s that the tool gives amateurs access to play where experts were required. So you are giving very powerful tools with out the requisite learning time of making mistakes or having other check and identify your mistakes.

I think that QA/QC processes will be challenged by AI and probably one of the larger risks.
GGG is offline   Reply With Quote
The Following 3 Users Say Thank You to GGG For This Useful Post:
Old 04-23-2023, 01:18 PM   #320
Itse
Franchise Player
 
Itse's Avatar
 
Join Date: May 2004
Location: Helsinki, Finland
Exp:
Default

For me, the problem isn't tools. The problem is that AI's (as they are) are most useful when asked to produce large quantities of something of unknown quality, fast, or when asked to create tons of new art, or when imitating human behavior.

These aren't really things we're short of.

We also aren't set up in a way where people working more effectively will have to work less or get to earn more. People are more productive than ever, but people are still working themselves to death, and only the rich are getting richer.

The speed and ease at which Firebot can create a functioning app only means that pretty soon every app store will be flooded and the quality of the app will essentially not matter. (This is already basically where mobile gaming is at. Having a good game doesn't really matter.)

Last edited by Itse; 04-23-2023 at 01:21 PM.
Itse is offline   Reply With Quote
Reply

Tags
they will overtake us


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


All times are GMT -6. The time now is 09:52 AM.

Calgary Flames
2024-25




Powered by vBulletin® Version 3.8.4
Copyright ©2000 - 2025, Jelsoft Enterprises Ltd.
Copyright Calgarypuck 2021 | See Our Privacy Policy