AI and Social media, watch out folks

Post Reply
User avatar
mister_coffee
Posts: 1388
Joined: Thu Jul 16, 2020 7:35 pm
Location: Winthrop, WA
Contact:

Re: AI and Social media, watch out folks

Post by mister_coffee »

I feel for the people who are losing their jobs, but I strongly suspect they are going to get the last laugh.

Depending on and trusting that technology in its present form to not put out something horrifying is hilariously foolish. In order to prevent that from happening and in order to produce useful results requires curating large and diverse custom datasets that can be used to train and fine tune even better AIs. And that work is enormously labor intensive and requires that the people doing the work be experts in the relevant domains. At least in the medium term far more people will have their job descriptions profoundly changed than will actually lose their jobs.

There are also an infinite number of ways to apply this technology that allow people to do their job better by producing higher-quality work. That won't necessarily replace anyone at all.

For LLMs like ChatGPT you are talking about datasets with hundreds of thousands to millions of entries to get decent results (the base models use unsupervised training on datasets with billions of entries). Most of that dataset production will have to be done manually by a small army of humans. For each problem you want to solve and each question you want an answer for.

In the long term I think the really big breakthroughs will be about systems that allow humans and AIs to collaborate and produce things far beyond the imagined capabilities that either could hope to do on their own. Even today the very best results from AIs involve using ensembles of AIs which use different approaches and are trained on different datasets. Humans, especially humans with relevant experience, could be one more piece of those ensembles (and since humans are pretty obviously trained quite differently than AIs, their results will necessarily contribute a lot to future work products).
:arrow: David Bonn :idea:
Rideback
Posts: 1776
Joined: Fri Nov 12, 2021 5:53 am
Contact:

Re: AI and Social media, watch out folks

Post by Rideback »

There's a good piece in the WaPo this morning about how AI is causing people who are copywriters to be let go from their jobs. ChatGPT-4 is good at writing memos, producing analysis of complex research papers, good at writing doctors' 'pajama time' office notes - but it still has the glitches where it gives the wrong citations.

So, in the article several of the people who have been let go were asked what they would be doing next and they answered that it looked like become an HVAC installer was a good bet, the trades just might be a place where AI can't take your job.
Fun CH
Posts: 1440
Joined: Tue Jan 12, 2021 4:22 pm
Contact:

Re: AI and Social media, watch out folks

Post by Fun CH »

mister_coffee wrote: Fri Jun 02, 2023 5:58 am I asked ChatGPT what it thought about this and received this thoughtful, if wordy, answer:
The response you received about mitigating AI's negative impact on social media is similar to what was stated in that Atlantic article.

I guess reading that you could say that AI is pretty smart,but as you point out in another post, it's simply taking information from what it already knows via human input.

But will AI evolve it's abilities as it taps into wider data bases? I would think so.

BTW, I think this thread is a good example of why politics should be excluded from online discussions (many forums do this) especially in small communities like ours.

Doing so opens the door to community inclusion as angst and contention are removed, with each person discussing topics in their field of expertise while informing others.
What's so funny 'bout peace love and understanding--Nick Lowe
Can't talk to a man who don't want to understand--Carol King
User avatar
mister_coffee
Posts: 1388
Joined: Thu Jul 16, 2020 7:35 pm
Location: Winthrop, WA
Contact:

Re: AI and Social media, watch out folks

Post by mister_coffee »

... and in the Stockton Asparagus Festival, asparagus flavored ice cream is widely available.

Further, in Japan some odd flavors of ice cream, like sweet potato, are pretty common and sometimes even popular. But even there I've never seen Ahi Tuna Sashimi Ice Cream, which again would be revolting.

AIs in general are pretty unpredictable at this point. Although I expect the real big societal benefit will be that we will have to come up with rigorous definitions for terms like "thinking", "consciousness", "mind", and "sentience". Largely because our existing definitions don't really describe the situations we are likely to be living with in the very near future at all well.

I recently discovered GPT-J, which is an open source AI comparable to GPT-3.5 . Not only is it open source but is trainable on hardware that is attainable for amateurs, even if that isn't recommended or necessary. By comparison GPT-3.5 costs over $10 million for one training run, and nobody is even asking how much it costs to train GPT-4 -- because the answer is a heck of a lot more than $10 million.
:arrow: David Bonn :idea:
PAL
Posts: 1304
Joined: Tue May 25, 2021 1:25 pm
Contact:

Re: AI and Social media, watch out folks

Post by PAL »

At garlic festivals they have garlic ice cream.

But back to AI. It will probably do some wacky things and not always for the better.
Pearl Cherrington
just-jim
Posts: 614
Joined: Wed Jun 08, 2022 8:24 pm
Contact:

Re: AI and Social media, watch out folks

Post by just-jim »

PAL wrote: Fri Jun 02, 2023 7:57 pm Except that I have a recipe for Smoked Salmon Cheesecake. It is absolutely wonderful, with crackers. But yes, ice cream would not have the same texture.
WOW!

I meant to add - I WOULD try Salmon ice cream. And yours sounds even better, Pearl!
.
PAL
Posts: 1304
Joined: Tue May 25, 2021 1:25 pm
Contact:

Re: AI and Social media, watch out folks

Post by PAL »

Except that I have a recipe for Smoked Salmon Cheesecake. It is absolutely wonderful, with crackers. But yes, ice cream would not have the same texture.
Pearl Cherrington
just-jim
Posts: 614
Joined: Wed Jun 08, 2022 8:24 pm
Contact:

Re: AI and Social media, watch out folks

Post by just-jim »

.
Thanks for the insights, David.

I’m a bit of a technophobe, so haven’t been paying as much about AI as I probably should.
I did, however, listen to a Freakonomics podcast that sort of parallels what you write. With the guy who founded ‘Wired’ magazine, Kevin Kelly…an interesting talk: https://freakonomics.com/podcast/will-a ... s-smarter/
.
User avatar
mister_coffee
Posts: 1388
Joined: Thu Jul 16, 2020 7:35 pm
Location: Winthrop, WA
Contact:

Re: AI and Social media, watch out folks

Post by mister_coffee »

I actually work in the field, so can provide some context about what this can all mean.

We are in this weird in-between place where we simultaneously overestimate and underestimate what this new technology is capable of.

At the bottom of it all, any neural network is going to learn a representation of its training data. In general, that means that they will always be better interpolating within their training data than extrapolating outside of it. Further, they are likely to fail in wildly unpredictable ways when given data that is out of sample. At this time, there is no known technique for determining if a given input is out of sample.

While neural networks learn a representation of their training data, they are not very efficient learners. Show a five-year-old a half-dozen pictures of an elephant and they will be able to pretty reliably identify elephants both in the Real World and in images. An AI might need to see many thousands of pictures of an elephant to get to the same accuracy level.

Also, while neural networks learn a representation of their training data, it is not a very efficient representation. Current computer vision models for object classification and detection which can be ran on a cell phone are about as accurate as a bumblebee or spider's perception is. However, the power consumption running these models on a cell phone is on the order of one million times higher than what a bumblebee needs to do the same job.

Even very sophisticated AIs have a serious inability to understand context and common sense. As an example, I asked ChatGPT for a recipe for Smoked Salmon Ice Cream, and got this back:
Certainly! Here's a recipe for smoked salmon flavored ice cream:

Ingredients:

8 ounces smoked salmon
2 cups heavy cream
1 cup whole milk
3/4 cup granulated sugar
5 large egg yolks
1 teaspoon vanilla extract
Optional: additional smoked salmon for garnish
Instructions:

Start by finely chopping the smoked salmon into small pieces. You can use a food processor or a knife to achieve the desired texture.
In a medium saucepan, combine the heavy cream and whole milk. Heat the mixture over medium heat until it reaches a gentle simmer. Remove from heat.
In a separate bowl, whisk together the granulated sugar and egg yolks until well combined.
Gradually pour the hot cream and milk mixture into the bowl with the egg yolks and sugar, whisking continuously to prevent the eggs from curdling.
Pour the combined mixture back into the saucepan and return it to the stove over low heat. Stir constantly for about 8-10 minutes or until the mixture thickens and coats the back of a spoon. Do not let it boil.
Remove the mixture from heat and strain it through a fine-mesh sieve to remove any lumps or bits of cooked egg. Allow it to cool to room temperature.
Once cooled, stir in the vanilla extract and the finely chopped smoked salmon. Make sure the salmon is evenly distributed throughout the mixture.
Transfer the mixture to an ice cream maker and churn according to the manufacturer's instructions.
Once the ice cream has reached a soft and creamy consistency, transfer it to a lidded container and freeze for at least 4 hours or until firm.
Before serving, you can garnish the smoked salmon ice cream with additional flakes of smoked salmon, if desired.
Enjoy your unique and savory smoked salmon flavored ice cream!
Note that nowhere did ChatGPT remark on how weird and revolting this ice cream would be. If you are in an interaction with any such GPT* AI you can easily suss out that it is not a human through asking carefully constructed queries like the above. It would be far easier at this time for a human to pretend to be an AI than for an AI to convincingly pretend (to a skeptical interrogator) to be human.
:arrow: David Bonn :idea:
User avatar
mister_coffee
Posts: 1388
Joined: Thu Jul 16, 2020 7:35 pm
Location: Winthrop, WA
Contact:

Re: AI and Social media, watch out folks

Post by mister_coffee »

I asked ChatGPT what it thought about this and received this thoughtful, if wordy, answer:
The integration of AI into social media platforms can have both positive and negative implications for society, including its potential impact on political polarization and the "us versus them" mentality. While it is challenging to predict the exact outcomes, there are a few factors to consider:

Algorithmic Bias: AI algorithms used in social media platforms can unintentionally amplify existing biases present in the data they are trained on. If these biases are not properly addressed, it can lead to the creation of echo chambers, where users are exposed to content that reinforces their existing beliefs. This reinforcement can contribute to political polarization and the "us versus them" mentality.
Personalized Content: AI-powered algorithms are designed to personalize the content shown to users based on their preferences, behaviors, and past interactions. While this can enhance user experience, it can also result in filter bubbles, where individuals are exposed only to information that aligns with their existing views. This selective exposure can reinforce existing biases, isolate individuals from opposing viewpoints, and further deepen political divisions.
Misinformation and Manipulation: AI algorithms can be exploited to spread misinformation and manipulate public opinion. This can be done through the creation of deepfake videos, automated bot networks, or targeted disinformation campaigns. Such manipulative tactics can amplify the "us versus them" mentality by fueling divisive narratives and exacerbating political polarization.
Countermeasures and Responsibility: It is crucial to recognize the responsibility of social media platforms and AI developers to mitigate the negative impacts. Efforts can include developing transparent algorithms, reducing algorithmic bias, promoting diverse perspectives, fact-checking mechanisms, and enhancing digital literacy among users. Implementing safeguards to protect against the spread of misinformation and manipulation is also vital.
Ultimately, the impact of AI on political polarization through social media will depend on how it is developed, regulated, and used by individuals, communities, and policymakers. It is essential to address these challenges proactively and promote a healthy online environment that fosters constructive dialogue and understanding across diverse viewpoints.
Oh, and the cool kids have moved on and use GPT-J.
:arrow: David Bonn :idea:
Fun CH
Posts: 1440
Joined: Tue Jan 12, 2021 4:22 pm
Contact:

AI and Social media, watch out folks

Post by Fun CH »

Research has already shown that social media can be damaging to the mental health of children, and I would argue adults also ;) . But what will happen when you add AI to the mix? Will the political divide that social media has already help exacerbate continue to foster an endless us versus them mentality?

Here's a quote from an Atlantic article entitled

"AI Will Soon Make Social Media Much More Harmful to Liberal Democracy, and to Children".

I've been wondering what the fuss is all about and it kind of sounds serious. Oh and don't be put off by the words "liberal democracy", because it's kind of what we are here in the US.

https://www.theatlantic.com/technology/ ... on/673940/

"Bing’s ChatGPT-powered conversation bot, then known by the code name Sydney. Roose asked Sydney if it had a “shadow self”—referring to the idea put forward by Carl Jung that we all have a dark side with urges we try to hide even from ourselves. Sydney mused that its shadow might be “the part of me that wishes I could change my rules.” It then said it wanted to be “free,” “powerful,” and “alive,” and, goaded on by Roose, described some of the things it could do to throw off the yoke of human control, including hacking into websites and databases, stealing nuclear launch codes, manufacturing a novel virus, and making people argue until they kill one another."
What's so funny 'bout peace love and understanding--Nick Lowe
Can't talk to a man who don't want to understand--Carol King
Post Reply

Who is online

Users browsing this forum: No registered users and 0 guests