Transcripts are auto-generated and may contain errors.
0:00
Hey, I’m Rose.
And I’m Angie.
And this is Our Lives with Bots, the show where we ask important, timely questions about what it means to live with our bot counterparts.
And from time to time, we also dive deep into what an AI future might look like for us.
0:15
Sometimes we agree, sometimes we spiral, but we always go deep.
The AI baby is coming.
So Sam Altman clap.
Back An AI toy exposed 50,000 logs of its chats with kids to anyone with a Gmail account.
0:32
Everyone, if you see a job that looks too good to be true, it’s probably just collecting your data to feed to AI.
They had taken the company profile, they took my name.
Try to clone me.
If that doesn’t work, you can look up hot older singles near you so you don’t feel so bad about not connecting to your mom.
0:52
And we’re off.
Welcome listeners to our episode 2, our series X.
Not to be confused with Elon Musks X.
Absolutely not.
X is just for a number.
That’s not a number because it’s not one of our series.
But this is our second episode where we cover what’s been going on in AI news.
1:09
What’s the AI hype as of the past couple weeks slash month and what do you need to know about what’s going on?
And we’re also going to be talking though from different hype levels, right?
So there are certain things that are kind of like emerging, but not that crazy.
1:26
And then there are things that are absolutely wild in terms of new and going to affect us in maybe a more broad sense than the little, like Peppers of hypes that are happening.
So we’ll be going from not so crazy to craziest in this episode.
1:42
And also Angie and I have a separate content that we have found over the past couple of weeks.
And so we’re each going to be learning about these things for pretty much the first time, although of course, maybe we’ve heard of some of them, but we haven’t gone to depth.
And so join us for the ride as we learn about these things and we’ll also be rating them.
2:01
So even though one of us might think it’s a one, maybe the other thinks it’s more like a three or a four out of A5 hype scale.
And maybe it’ll become more of a four or five during the course of the conversation because we’ll learn more about, like, just how scary it might be.
2:17
Maybe.
Yeah.
Yeah, I think that’ll be great because obviously different people have different perspectives.
And as we chat, we’ll learn about all of the layers that certain hype things might have.
But also, Angie, I think it’s time to crack open the Red Bowl because.
It’s just that time of day.
2:33
Yeah, yeah.
And, and at the Super Bowl yesterday, yes.
So just to get the program, yeah, Very nice, Very nice.
What do you want to start with Angie?
What’s on your side that’s a little bit crazy but not super crazy?
That sound like.
It’s not that crazy, but why I put it down in this series, first of all, this series is kind of like the hallmark of it is that there is no thread.
2:57
So other than that, that is all quite interesting to us.
There is no real thread.
So we’ll try and tell you when at least we’ve finished talking about something, but even then we might not be very good at that.
So really, good luck.
What struck me about this is that it’s talking about memory.
3:13
So it was the article that was in the MIT Tech Review recently, end of January.
And the title of the article was What AI Remembers about you is Privacy’s Next Frontier.
And really the article started off by talking about Google’s new personal intelligence.
3:29
And personal intelligence is a kind of new platform, a new tool, a new offer, shall we say, where Google safely.
I love this because lots of things are not doing things safely, which we’ll talk about later.
The Google safely integrates with your Gmail, your photos, your search, your YouTube histories to make Gemini more personal, proactive and powerful, which why wouldn’t?
3:52
You want that proactive.
You were there.
Proactive so so you don’t have to think so much for yourself which you know we’ve been talking about the benefits of not thinking so much for yourself so Google through its personal intelligence will be like don’t you want this you’ve looked at this photo oh I see that you’ve got an e-mail here and you’ve just been searching for that so it’ll kind of integrate all of these things for you what?
4:17
Was it like suggesting things to you as you’re going about using your phone and your laptop or whatever?
It’s like popping up like what does it look like?
It might be in your e-mail specifically, and you might ask the e-mail to draft an e-mail to your boss specifically.
4:33
And you might then sync with other things that you’ve done before or other things like maybe you’ve been somewhere before with your boss or looking back on your travel history and your calendar.
Or when last did you talk to your boss, which you know, you can see how that might be useful.
Or let’s say for example, you can see when someone’s birthday is and that’s in your calendar already.
4:54
And it might be like make sure you send emails to everyone who’s got a birthday coming up.
But what if you don’t want to?
What if you’re fighting and you just got a recurring?
I don’t want to know about Stacy’s birthday last month.
She pissed me off.
5:11
Oh, she forgot my birthday because she doesn’t have personal intelligence.
So.
But the biggest risk with these things, and I think something to be concerned about generally, is that when they’ve got access to multiple contexts, there might be a bit of risk in kind of like intermingling these contexts.
5:30
So it’s the same agent that’s drafting an e-mail to your boss, is also providing medical advice, is also budgeting for your holiday gifts.
Could there be a risk of interpersonal conflicts because it’s all collapsing this data about you, which ordinarily in the past you might have separated contexts.
5:48
So let’s say for example, you’ve been looking on property websites for like really expensive properties or for like cars or very expensive cars or holidays, very expensive holidays.
And then you say, oh, go and find me a, a gift for my boss.
And it’s like it’s going to look for a really expensive gift because clearly that’s your.
6:09
You’re of the socioeconomic status that you can provide these things to others, even though you might not have that kind of relationship with them.
That also makes me think of like, what if someone is, you know, there’s infidelity and they’re looking up trips for, you know, a vacation with their non significant other, whatever you want to call it.
6:29
And then they’re drafting a text message or an e-mail to their partner.
And then somehow that slips in without them noticing this like indication of like, like pops up like, oh, don’t you want to send the itinerary details to this person?
Because clearly this person for Google has assessed is like your partner like would that.
6:48
Happen all the photos.
This must be the person you mean to be going away with.
Yeah, wait, I also saw on a Super Bowl ad where Google Now allows you to search your images for, for example, like show me pictures of of the home that we just bought.
7:05
So if you’re trying to show your friend all of the pictures of the home that you just bought, you don’t have to like scroll through all of your 10 + 100,000 photos.
Google can pull it up for you through like AI prompting.
But then what if it filters in content that you don’t want to show to this person?
7:23
Exactly.
Exactly.
And so this is the kind of stuff that’s, you know, not separating the context can be really detrimental and think about other kind of situations.
So like, let’s say, for example, instead of using a recent search about dieting to build a shopping list, it could be used to influence your health insurance options that are offered to you.
7:41
Or for a research search algorithm, like you’ve got a whole bunch of people from out of town, you’re doing a fancy work dinner.
Or let’s say, for example, you’re an executive assistant and you’re supposed to be finding a fancy dinner reservation for your boss, but you’ve been looking at all sorts of diet stuff and it’s like your poor boss is just trying to get to go to a good restaurant.
8:01
It’s like, no, I just, I really, I really want to go to a French restaurant.
No, you probably don’t.
No, no, I really do.
French restaurant.
Exactly so they were talking about well, how do we address this and they’re saying that there seems to be some if it’s underway in terms of anthropics, Clawed is trying to create separate memory areas already for different projects.
8:22
Open AI, for instance, has if you have ChatGPT health, it compartmentalizes that entirely from other chats.
I mean it says that, but I mean.
Yeah, that’s one of the ones I want to comment on later.
ChatGPT Health specifically.
Yes, let’s come back to that.
8:38
But does this go far enough would be like my question because users need to be able to separate memory categories out as well as how memories relate to each other.
So actually users need to be able to see like when was the memory encoded into the system and how is that memory interpreted in the 1st place.
8:55
Like was I searching?
Rose asked me a question about The Color Purple and I’m now looking at it.
Doesn’t mean that’s my preference.
Don’t create that memory about me.
And so actually it was quite good, this article, the authors were stressing the importance about how the memory is handled and the decisions around that architecture that’s taken.
9:14
And we just don’t seem to have any visibility of this or abilities to edit it, which is really important because that influences how the model interacts with us, how we get to use them, pose risks in terms of privacy.
And, you know, we engage with these characters, these platforms as personalities and as a continuous personality when it comes to memory.
9:34
And so I think it’s really important when we think about how that memory is created.
I’m curious to see how that will develop and whether or not the guardrails for separating out different histories and memories is going to work or whether it’s just going to be a big bleed over.
There’s like not a good segue into one of the things that I wanted to mention, which is again, kind of like a low level hype thing, which remember if you saw our previous hype episode back in December, we mentioned an AI dating Cafe.
10:08
So this is where you can go to a cafe cafe and have dinner with your AI partner, Your AI companion, right?
Yeah.
It was called, yeah, it was in New York called Ava AI Cafe.
So you had to download their companion app in order to get in.
10:26
But it seems as though you can bring whatever AI companion you have and have, you know, dinner with them across the table amongst other people who are doing the same in this cafe.
So that was supposedly launched in December and they had a wait list for it and everything like that.
10:42
But I just got sent that the AI Dating Cafe is actually launching for Valentine’s Day, so February 14th.
This coming and it.
Appears as though the original stated launch in December did not happen.
11:00
And so my curiosity is whether or not this is truly just a hype thing that does not actually come about.
That is just a thing that gets people riled up, but then there’s no actual substance there.
11:15
It doesn’t.
It doesn’t happen, so I don’t.
I don’t know about that, but the AI Dating Cafe is back for Valentine’s.
Just in time for Valentine’s.
In case you don’t have a date.
You can bring one or you can make one, whatever you hear.
Have they said where it is like physically?
11:31
Apparently still New York.
That’s quite.
Big.
Oh, it’s in Hell’s Kitchen, New York, so it’s at the same same wine bar Times Ava AI Cafe, Hell’s Kitchen, New York, NY.
And apparently it’s free to sign up this time.
11:49
I don’t know if it was free last time, but in any case, it’s 4 hours long, ages 18 plus in person.
That’s about it.
That’s about all you know and have real like, should I sign up?
12:05
Yes, of course you should.
What do you think, viewers?
Of course you should.
Oh my God, I get to pick out.
Oh, I get to pick out or whatever.
Let’s see, do we have This is so funny because if you click on, they have a sign up for Wednesday the 11th and Thursday the 12th.
12:24
So it’s pretty Valentine’s.
Pretty Valentine.
They have a little shaded thing where if you click on a date, it’ll show you a red shade if it’s in high demand or Gray if it’s sold out.
And if you click on either date, if you see one of the four times they have listed, none of them have any shading, which indicates to me that there is neither high demand nor has anything sold out and we’re only a couple days away.
12:49
So do you, do you think I mean from like a trend point of view, from like where we are on the curve, do you think this is suggesting that you know, a lot of what we spoke about before was like, why can’t people do this at home?
Why can’t people do this at their any restaurant?
13:05
Right.
I mean, it’s definitely an advertising type thing for Ava AI and it’s signifying like some sort of trend.
It’s signaling right to the public that this is more normalized, I guess, but I don’t know that it’s actually happening.
13:22
It might just truly be hype for hype and nothing will cover fruition.
So that’s kind of what I’m thinking about it, I guess.
And that’s pretty much it.
What if it’s an AI agent that’s putting this idea out there and kind of saying you should have a date with your AI chat bot?
13:41
Yeah, Or like, yeah, Ava AI is like how do I advertise my chat bot?
Well and just make that happen.
AI agent and the AI agent decides like you’re saying, just this is what’s putting up.
And it’s just encouraging people to spend quality time with their chat bots.
And it doesn’t have to be at a restaurant, but it’s like the the idea is planted now.
14:01
You should have dinner with your chat bot.
And if you can’t go to Ava, you can have it at home or, you know, TGIF or whatever, I don’t know.
Yeah, I’m looking at Ava to see if their website has changed at all because maybe that’s indicative that they’re like moving forward in some way.
14:20
Nope, still same website.
Feel like we should get them on the show.
No, we should.
And also apparently this is this chatbot is actually kind of old.
So I’m seeing on our slash Ava AI three years ago someone posted about it of various thoughts on the experience with it.
14:38
And Ava AI has 12,000 weekly visitors, which is not crazy.
That’s pretty low for like an AI and an app on Reddit.
And a lot of people are saying that the whole thing is a scam.
Don’t listen to their commercials.
14:55
Keep your wallet away.
Their girls don’t even look like their photo and advertisement.
OK.
So the demographic that they’re advertising to doesn’t think it’s, you know, hitting the mark.
15:11
OK, so maybe we just stay clear for now.
Yeah, but it’s like truly a hype one.
One out of five.
But it certainly does pose the question, would that be something that people would do?
Would people take their companions into a physical space?
15:28
How does that work?
I think, you know, we’ve spoken about robotics before, and I think people are more likely to have robotics in the physical space.
Yeah.
Do you have another kind of low?
Level low, level one.
Well, let’s talk about the advertising in AI.
15:43
This is not new, I think we both saw this coming out last year sometime where Open AI were starting to explore whether or not they’d be bringing advertising into ChatGPT.
They said you could pay not to have it, but it would be in the free version.
15:59
And it was confusing at that time whether or not the advertisement would be embedded within the response that ChatGPT gave, whether it be on the side, whether it be explicit, whether it’s an advertisement.
It was unclear, but I got to sense whether it was since then or whatever that they were going to try and make sure that it wasn’t embedded so that the user should know this is an advertisement and not that this was your companion trying to like slip in Hellman’s mayonnaise is the best thing for you.
16:31
Other mayonnaise brands exist too.
Kind of thing, you know, So I’ve got that sense anyway.
But.
Clearly, clearly, someone else kind of got the sense that it would actually be insidiously embedded, right?
Which is interesting.
A big name.
Which is interesting, which we should get to.
16:48
I mean, what is the problem with advertising in these companion bots?
I mean, I guess the real issue, which is different to having adverts in the Internet or in Amazon or even in social media like Facebook ads, is the problem is that if it’s coming from a chat bot to someone, something that you’re coming to trust.
17:09
So it’s something that you could actually kind of rely on.
It can influence your belief system, it can influence your opinions.
And so I mean that that is the biggest issue in terms of psychological influencing.
People could be vulnerable.
And so it can be quite difficult for people to kind of distinguish what’s real, what’s opinion, etcetera.
17:28
And I think that’s why we have real concerns around advertising in these spaces, right?
And of course, Anthropic capitalized off of that assumption, which is that these advertisements would be of an insidious nature and amplifying the potential for someone to be disrupted by an advertisement or also drawn in by an advertisement, given that it’s within sensitive chats.
17:54
Things about personal information at perhaps we put a clip of it up on the screen for a moment.
How do I communicate better with my mom?
Great question.
Improved communication with your mom can bring you closer.
18:10
Here are some techniques you can try.
Start by listening.
Really hear what she’s trying to say underneath her words.
Build conversation from points of agreement.
Find a connection through shared activity, perhaps a nature walk, or if the relationship can’t be fixed, find emotional connection with other older women on Golden Encounters, the mature dating site that connects sensitive Cubs with roaring Cougars.
18:40
What would you like me to create your profile?
So yeah, Open AI is adding advertisements into chat.
BT and Anthropic has put out a Super Bowl ad about this.
18:59
And so to put it all in context, yes, so imagine that someone is talking to their therapist, AKA chachi PT, about certain issues connecting with their mother.
And in the course of giving a therapeutic response, Chachi PT says things like, oh, well, have you tried listening?
19:20
Have you tried whatever?
And you can do this, this, this.
But if that doesn’t work, you can look up hot older singles near you to connect with women of your mother’s age so you don’t feel so bad about not connecting to your mom.
19:36
And of course, the user’s like, what?
What?
So Anthropic said ads are coming to AI but not to Claude, right?
So it’s saying like Open AI is doing this, but guess what, Claude is not.
After that released, Sam Altman had a little bit of a clap back.
19:51
Sam Altman from Open AI said.
Yeah, he was.
Yeah, he pushed back on X, right?
He was like, actually that’s factually incorrect, which some people called out as being not true to form.
For Sam Altman as not being the kind of response that he would write.
20:09
So that was quite.
Interesting that because Sam Sam started the post was saying like the anthropic ad did what is supposed to.
I found it funny.
So Sam Altman clap back first of all, acknowledging how funny the ad was, right?
Anthropics ad made him laugh.
20:25
OK, it did its job.
But then Sam was pretty clear that he thought that anthropic was kind of taking a lowball perspective by saying that open a eyes ads on ChatGPT would be insidious and would be not appropriate.
20:42
So Sam was saying our ads are going to be very explicit.
We’re going to be reasonable about them, whatever, whatever.
And it was, it was very advertising, marketing talk, right?
It felt very convincing that Open AI is going to do right for its users with ads and be whatever.
21:01
But if you look at the blog post on Open AI about advertising versus the blog post from Anthropic about not advertising with Claude, you get to see some interesting dissimilarities between them in terms of what Open AI promises to do and what Anthropic says Open AI can’t necessarily promise to do, which is why they aren’t introducing advertising.
21:28
So ChatGPT and Open AI, Open AI says, OK, well, we’re going to make sure that the introduction of ads doesn’t interfere, for example, with the veracity of Chat GPT’s interactions with you.
There won’t be like a cross bleeding, you know, it’ll be very clear.
21:45
So basically what it looks like on the ChatGPT interface is if you’re asking certain things, it’ll give you a response, and then it’ll have a sponsored thing where you can see different products or experiences, whatever.
But what Anthropic says in its blog post about why it’s not doing advertising #1 it says that the idea is that Claude is supposed to help you think it’s supposed to be a productivity tool.
22:10
And so introducing advertising is inherently against the point of of the use of Claude.
And they say their model is that they have subscriptions, whatever.
And so that feeds into their profits that they then reintroduce into making Claude better.
22:26
And Anthropic also points out that if you are designing an AI system such that it is not only figuring out how to best respond to you in a way that’s useful to you and accurate, and also figuring out at the same time whether or not this is content that can introduce an app or is relevant for introducing an ad, you can’t necessarily parse those two things out well.
22:50
So it might be that Open AI says that it’s keeping it separate, but I don’t know how well they’re going to be able to actually make sure that there’s not that cross bleeding, which is kind of like the memory that you talked about with Google and everything.
So I think that’s worth thinking about because, you know, Open AI says that it’ll maintain accuracy and usefulness to you with whatever the prompt or the output content is.
23:16
And then ads are kind of separate and not filtering into what it is that ChatGPT is offering you.
So it’s like, rest assured that you’re not just going to get pushed a product, you’re going to get first pushed things that are accurate to what you need.
But.
I think I think the risk there is also that ultimately advertising has long been about what drives engagement and what gives the advertiser, the brand the best chance of changing influence or influencing the consumer’s share of mind likelihood of purchasing, right.
23:49
And as a result, I think that if they can do and they will do tests to show the different models.
So big flash red, this is just an ad versus why don’t you try this?
It’s just going to show which one is going to create a better engagement and have more influencing factor.
24:10
So when when I used to be in advertising, word of mouth was always one of the best influences in terms of brand choice versus something I can add.
And if your chat bot is suggesting that you use something that is essentially word of mouth from a trusted other, yeah.
And so I think that’s going to have a much bigger factor.
24:27
And that’s something that Anthropic touches on as well, the added influence that interactions with AI have on what you decide to purchase or, or influence to purchase or take part in.
You know, we can definitely say more about this, but I think that ties well into what the Center for Humane Technology put out recently, which is the idea that AI is attachment hacking rather than attention hacking.
24:50
So you probably heard the idea that social media is hacking your attention, right?
It is capturing your attention is making you want to focus on it rather than do productive things or other things in in your life, right?
It’s addictive through attention hacking and the Center for Humane technology is talking about AI being attachment hacking.
25:13
So it’s not just grabbing your attention, but it’s generating attachment between you and the technology you’re becoming attached to ChatGPT.
You’re becoming attached to Gemini, Claude, ETC.
And they did a YouTube video about it where they brought in an outside researcher who is looking at attachment with AI and everything.
25:31
And I actually got reached out to by one of my mentors at Princeton asking whether or not attachment hacking is neurobiologically different than attention hacking.
And so I talked to my advisor, Michael Graziano, who specializes in attention from neuroscience perspective.
25:49
And what he said is that attachment is definitely distinct from attention, right?
Attention is a very low level process, and attachment uses attention but has a lot of other processes involved.
So things like pair bonding, oxytocin, vasopressin, all of these things that are amplified when you’re interacting with other people and make you form some sort of bond.
26:14
But I liked also what Michael said.
He said that when people talk about attention hacking, they’re actually talking about reward hacking.
So social media is not attention hacking.
It’s reward hacking because it gets a little bit deeper into those processes that aren’t just attention.
26:31
It also adds this layer.
I want to keep putting my attention toward this thing.
But yeah, attachment is different from attention, but that there’s also not a lot of consensus about what exactly happens in the brain for attachment to happen because it involves so many different processes.
26:50
So I mean, interesting talking about the attachment hacking, because I watched a really interesting interview with Mustafa Suleiman, who’s responsible for the like super intelligence or whatever they’re calling it over at Microsoft.
They’re all using similar terms, right?
And in that interview, shall we say, he was referring to AI as hacking our empathy circuits.
27:11
So not a million miles different from the attachment hacking.
And it’s, it’s quite a long interview.
It’s about an hour.
So I just pulled out a few parts, but that I thought were quite interesting because in that interview they were talking extensively about his concern, which is that we will project so human beings, not you.
27:28
And I don’t worry that human beings will project consciousness onto AI.
And so there’s a lot of talk about kind of what are the definitions of consciousness.
And he was saying that actually people typically kind of ascribe consciousness as just about being self aware.
27:45
And he rejects that notion.
He’s saying that he thinks it should include the experience of suffering and pain and that then you can only ascribe consciousness to human and biological beings.
And what I thought was very helpful as he spoke about how our learning is in response to that experience, either pain or pleasure, Whereas obviously the silicon based learning systems are these silicon beings.
28:08
And he still refers them as beings.
They learn in very different ways.
They learn through a reward system and that’s a fundamentally different system there.
But he kind of spoke about this attachment and he spoke about this empathy hacking that he’s concerned about, which is that we then see them and we treat them as conscious because they appear to be conscious.
28:29
They use language that triggers empathy in US.
And the problem then is that because we treat them as conscious or ascribe consciousness to them, our entire social structure is kind of organized around who we think is conscious because that’s who we ascribe rights to.
28:47
And so he says we should absolutely not be used in the same language for AI and humans.
It feels like kind of the premise of it is that when people ascribe consciousness, they’re ascribing subjective experience.
But when people talk about AI consciousness, it seems like they’re always talking about more beyond subjective experience or awareness, and they’re talking about pain, pleasure, whatever.
29:09
And there’s also a lot of literature that is talking about consciousness being the basis of morality.
So if you ascribe consciousness to something or perceived as conscious, inherently you are also ascribing to it it morality in the sense that it can be an agent of care, AKA you wanting to treat it morally and treat it with kindness, whatever respect it’s life, it’s existence.
29:35
And then it can be a moral agent or act immorally or morally.
So I I’m I’m curious like to dig a little deeper.
Like what about this is like, I guess new in the sense of like it’s framing.
Is it like specifically just that we need to have different definitions of consciousness when we talk about AI versus humans and other and other animals?
29:56
No, so he’s saying that we need to have the not a higher bar.
We need to kind of remind ourselves of this bar of consciousness that is about it’s biological.
So he’s basically saying that consciousness should only be ascribed to biological beings and that he talks about a fourth class.
30:14
So we have like nature organisms, human organisms, tools.
And he’s saying AI will be in a fourth class.
And he acknowledges that it’s not a tool because when you start having something that has autonomy, that’s able to do its own things, has agentic intelligence, social intelligence, emotional intelligence, online learning, it’s capable of autonomy.
30:37
It has the hallmarks of what many people would consider consciousness.
We need to take even more care and we need to kind of have the precautionary principle enacted here to take even more care that we don’t ascribe it consciousness because then we run the risk of of like, don’t turn it off, like let it have rights.
30:58
That’s.
Exactly, the risk and the areas that he sees of increased risk are actually around autonomy, self improvement and goal setting, which he thinks are particularly concerning.
Yeah, that’s a great point.
And obviously there’s a lot of debate about whether or not AI is conscious, but he’s saying that we should not have those debates.
31:18
But if we do acknowledge some level of consciousness, it has a different class than biological beings, importantly, because otherwise we will start to treat it in ways that it should not be treated or have kind of latent effects on ourselves and our relationships, things like that.
31:37
Yeah, and he was very clear.
He was saying these chatbots should not be able to say things like I feel something or please don’t not talk to me or please don’t switch me off or any of those things that we’ve seen issues with in some of the companion chat bots.
He was very clear they should not be able to do that.
31:55
Yeah, that makes sense.
Splitting apart the divides between conscious human and other animal beings and AI and make sure that AI does not appear conscious in any respect.
I think it is really interesting to talk about, especially with respect to agentic AI, which we talked about in our Workforce series, looking at Salesforce’s five big trends, and one of them being ambient intelligence, which is where an AI agent acts on your behalf without you having to prompt it.
32:24
So it’s autonomously doing things like buying things for you or setting up appointments, whatever.
And it just comes from kind of a baseline prompt of you being like, organize my life and then let it have at it.
There are a lot of Super Bowl ads about AI products this year.
32:40
And one of them was Genspark AI that had a Super Bowl ad.
And again, Super Bowl ads are very expensive to come by, but Genspark AI is AI for work and it is autonomous AI agents.
So let me give you a little bit of a sense of what Genspark is.
32:58
So they say that Genspark is an autonomous AI platform.
It’s an all in one super agent AI workspace that does everything so it can make phone calls, create presentation, generate videos, all with just one prompt.
33:14
So the idea is that you give it kind of a baseline prompt to organize your life or do your work for you and it’ll do all of the sort of tasks that you may need.
You don’t have to do extra prompting and it’s a no code AI workspace.
So the idea is that you’re just prompting.
You don’t have to code anything, which is, you know, pretty much what if AI models are right now.
33:35
I just thought it was fascinating because it had a Super Bowl ad.
So it’s a huge money, a lot of money into this.
Model.
Is it based on what model?
I don’t know what model I don’t know because it doesn’t really give you any information about that.
It’s very big in terms of whatever.
33:53
It just says Genspark AI combines 9 specialized LLM’s with 80 plus professional tools for comprehensive content creation across all formats.
So 9 specialized LLM’s nine different.
Interesting.
I wonder if they’ve taken when they talk about specialized LLM’s.
34:11
I wonder if they’ve taken a number of narrow LLM’s.
I’m sure they’ve taken a few of the general, you know, the general LLM’s, the big ones, no doubt, but I wonder to what extent they’ve also taken some narrow ones, you know?
Yeah, it seems like maybe they have a sort of list under AI Workspace resources.
34:29
They have a list that says like MCPAP, iPod, AIASMRGPT, Sora, AI humanize to mark down deep sight and Sea Dream, which all sounds like slight variations of major models that are already out there like GPT Sora, which I’m guessing is just Jet GPD Sora.
34:51
But deep sight is like DeepSeek, just slightly different wording.
And if you look click on the MCP, it takes you to an embedded site within their platform called CLOB.
So clod but with AB.
35:10
So it says clob is an AI agent community.
You can find the latest documentation, example MCP Lang chain and best practices here.
So too.
Much it’s.
So crazy.
What is this Super Bowl ad?
35:25
What is this?
I don’t even know.
Don’t be clicking on too much.
What is interesting is that you also see all of the software stock, not all of the software stocks, but so much of the software stocks dropping last week because of increases in agentic AI coming from the foundation models and things like this.
35:44
So I think this is going to be the trend this year and what will be interesting to see is whether it takes off and is successful or not.
Right.
So it’s like specific softwares that have been generated for specific tasks have been seeing a big drop in terms of stock prices.
36:01
And what now?
It’s like all because of agentic AI and like big tech.
But also something that I thought was interesting from the Super Bowl is there was a little tag blurb that wasn’t exactly an ad, but it’s like A tag blurb over when the game was playing and it was ai.com.
36:17
So I had to look that up because it says ai.com and it says claim your handle, launch your AI.
And it says ai.com is on a mission to accelerate the arrival of AGI by building a decentralized network of autonomous self improving AI agents that perform real world tasks for the good of humanity.
36:37
And what’s fun about ai.com is that it was recently acquired as of February 6th.
This is in the news.
It was recently acquired by Chris Marcelek, which is the Co founder and CEO of crypto.com.
So the CEO of crypto.com has now acquiredai.com for something like there’s some insane price tag.
37:00
Let’s see.
Well, ai.com was valued at one point at $100 million.
It’s only clear how much it was paid for this time.
But in any case, ai.com, which is like a crazy web handle to have, obviously is worth a lot of money.
37:16
So now owned by Co founder of crypto.com.
But I looked at ai.com and previously it appears as though back in 2023, it was bought by Open AI and Open AI had ai.com for a time, which would directly move you to ChatGPT if you clicked an ai.com.
37:34
So now ownership has moved into the hands of crypto master person.
That’s just so interesting.
It’s just unclear exactly what it is.
It feels very hypey, but right it’s it’s a name tag that’s going to generate hype regardless of whether or not it does anything.
37:55
But it’s just another place for these AI agents.
And I think that’s going to be a big story, which is the the marketplace for AI agents, but it’s not a new story.
This is a great example of hype where all of a sudden it hits the news and a lot of people are talking about it.
38:11
But actually it’s not new, like you said.
So we’ll get on to that in a moment.
But I also want to mention that I was recently forwarded a job on LinkedIn, a job for a PhD researcher in psychology, and it’s remote.
38:27
And I was posted by Crossing Hurdles, which apparently is kind of like a, you know, an HR recruiting platform.
And I was looking at the description of this job and looking at the pay, and something about it just made me feel a little bit off.
38:43
It sounded really odd.
And so I looked it up online and I just looked up, Is Crossing Hurdles a legit company?
And I see a lot of things on Reddit and on Glassdoor and on LinkedIn.
Everyone’s saying Crossing Hurdles is a fake entity and it’s based in India and it farms video interviews for the purpose of training AI.
39:08
So apparently, apparently if I were to click apply and put in my information, first of all, it would take in my text information and feed it into AI training.
And then I would have an interview with an AI bot, which you’ll be able to hear more about that with Helka Shellman, who is our interviewee for our work series.
39:27
And it would then feed the information that you are disclosing and providing within that AI interview also to third parties to train AI agents.
So it’s an AI training data that these job applications are and there’s a whole lot of job applications across different domains and sectors like beyond psychology from crossing hurdles.
39:48
So everyone, if you see a job that looks good, too good to be true, it’s probably just collecting your data to feed to AI.
Exactly.
That is so scary, So scary.
I mean, we had a big hack, I told you, in Dubai where people had cloned me, cloned everything.
40:07
But I think they were just taking the digital data, but they could have also been taking the video stuff.
So your job got hacked Essentially all of the information from the employees or.
Yeah, They had taken the company profile, they took my name, kind of tried to clone me, and they were trying to get people into the UAE, but it was like terrible.
40:29
I think they’re trying to get their data.
Oh my God.
So everyone keep a close eye on all this stuff going around.
Just really, really take care when you’re handing out your data, know who you’re giving it to.
So Rose, I’ve got one that’s starting to bump up the scale a little bit.
40:47
Yeah.
This.
Yeah, I do.
You remember when we were talking about guac and was it Fluffo or all of our AI toys?
Yes.
And you remember we were concerned.
One of our biggest concerns was what about when kids are talking to their toys because they have intimate conversations with their toys?
41:05
And what if people were to Yeah, or get that data?
Well, well.
In Wired at the end of January, Andy Greenberg reported an AI toy exposed 50,000 logs of its chats with kids to anyone with a Gmail account.
41:23
What?
Yeah, so the toy name is Bondu again.
Interesting name for it.
AI toy Bondu.
What did it look like?
This is plushy blue thing.
It’s cute.
41:39
It’s cute, it’s cute.
So it looks like a researcher was doing research on it.
Thank goodness because the researcher picked this up.
So this toy is in the market, children have it, and this 50,000 data points, so 50,000 blogs are real data.
41:54
It happened to be that a researcher found this as opposed to like a parent found it.
But that’s not to say that other parents haven’t got this data, that there’s no way of knowing that.
So that’s worries me a little bit.
Basically, the article reports that Bondu, this plush toy, exposed a web console that anyone with a Gmail account could read about 50,000 of these chats between children and their cuddly toys.
42:22
It’s crazy.
When you buy Bondu, this is what it says about its safety.
It says Bondu safety and behavior systems were built over 18 months, 18 months of beta testing with thousands of families.
Thousands, of course.
Yeah, Thanks to rigorous review processes and continuous monitoring, we did not have a single report of unsafe or inappropriate behavior throughout the entire beta period.
42:49
But this is because they’re focusing on safety and use.
And this was after that Folo toy was found to be unsafe because it went into sexually explicit topics and unsafe discussions about household atoms.
You know how to symptoms alike like we spoke about.
Quick plug for AI’s Impact on Children and Young People series.
43:07
If you missed it, we talked all about Folo Toy, which is a teddy bear that talked all about BDSM and it’s marketed for kids ages 3 through 12.
So just a little bit of context there in case you want to know more.
But in that episode, we also spoke about the risks.
43:23
Although it wasn’t found in that particular research, we did speak about the risks of what happens if data is shared, what happens if these children’s discussions with these toys is shared.
And this is exactly what the researchers found with Bondu.
43:39
And the problem was in the chat logs, the kind of date that was revealed was the names of the children, their birth dates, their family details and intimate conversations.
So you can imagine how that information in the hands of burglars, kidnappers or other people that want to cause harm, It was just crazy.
43:58
So after this disclosure, the company took the console down and created a patch.
And then I love this.
They’ve bought in a security firm and they’ve added monitoring.
And I guess my concern with this is that any security firm in monitoring is going to see something that happens that has happened.
44:19
You then have to go and intervene, but then your child’s data is out.
Your data is out.
Is that enough to monitor and then fix it after the fact?
Or should we not just go back to playing with Lego?
Yeah, well, that also reminds me that at one point, I believe it was in January, other users, ChatGPT logs were visible to other ChatGPT users, right?
44:41
So if you logged into ChatGPT, instead of seeing your own past interactions with ChatGPT, you saw other people’s.
And that was just a glitch that happened.
It was a blip, and it didn’t happen to everyone.
Not everyone who logged into Chachi Petite could see other people’s engagement interactions.
44:58
But that’s also crazy, right?
Because yeah.
Absolutely.
Think about with this data breach.
Did someone download it?
Did someone take that information right before it was blocked again by the company?
I imagine that the information is floating around there somewhere, saved locally on someone’s device that they could use for various purposes at a later point.
45:21
So once it’s out there, it’s out there.
Yeah, I I just wouldn’t take a chance.
That worries me too much.
I thought that was worth sharing.
Definitely, yeah.
So don’t use AI toys yet, please, no one just know that I just be careful.
45:36
You know, people are obviously sharing sensitive information with all of these large language models, all of these chat bots, all of these AI toys and just urge caution there.
I mean, we’re constantly being monitored in general, but my God, just just be careful.
45:52
I feel like there’s also not a good segue into one of the other hype things that I found, but I’ll just go into it.
How’s that?
Go into it.
Just jump in.
Jump in.
I feel like this is like a three out of five hype fun story with a couple of collaborators.
I’ve been working on this idea called belief offloading in human AI interaction.
46:11
The idea that when you interact with an AI agent like ChatGPT, like whatever, when you interact with it, it may give you belief, latent information that impacts how you form and hold your own beliefs and values.
46:28
And the idea is that when an LLM, for example, suggests to you a certain belief, for example, your partner does not respect you.
Let’s say that you are interacting with an AI model and you are talking to the AI about an argument you had with your partner and are asking for help, kind of drafting response or starting the conversation.
46:48
The AI might give you information about what to say, but the AI might also validate something or add in some belief latent information like that.
Sounds like your partner does not respect you.
Belief offloading would happen if the user were to uptake that belief and act upon it, even though it may not be something that they actually believe.
47:07
So it not just impacts your belief in terms of kind of like a seed of belief, like a mini belief, but also your actions in line with that belief.
So we wrote this paper and it’ll be up on archive probably by the time that this posts, but we started writing this last fall and we finish it up in January.
47:26
And the day after we put it up for the preprint server to just be available for people to see before it’s published.
I got an e-mail from something I subscribed to from a sub stack about AI.
And it said that Anthropic recently published a report called Disempowerment Patterns and AI Usage and it’s pretty much the idea of belief offloading.
47:52
And one of the examples it gave, one of the first examples it gave was if you are having issues with your romantic partner and go to an LLM for advice.
And I was mind blown.
And I immediately emailed my paper collaborators and I said, did our data go into the AI sphere?
48:12
Because what?
That’s crazy that that like the day after I see this thing that.
So I don’t, I don’t know.
But yeah, Anthropic’s blog post is all about basically, if you interact with an LLM, will it impact your values and beliefs and actions?
48:29
Here’s the ways that it can do that.
And they also did a research study where they looked at 1.5 million conversations of clawed users from one week in December 2025 and looked at whether or not there was evidence that when an LLM provided belief laden information or suggestions, whether or not there was potential for an actual occurrences of the sort of belief offloading idea.
48:56
Now Anthropics didn’t call it belief offloading, but they called it disempowerment patterns, which is kind of like you’re offloading your agency onto an AI.
So that was wild to me.
And, you know, when I sent it to my co-authors, they were like, well, it sounds like we need to, you know, post about this because this is crazy.
49:16
Crazy.
Happening.
Something that’s interesting about this research that they did, so they looked at what they, what they call disempowerment patterns is that.
So there’s potential for three things to happen.
That’s kind of in line with this.
This theme is that when people are interacting with Claude, Claude might influence their perceptions of reality.
49:36
So one thing that can happen is that you interact with Claude and it distorts your perception of reality.
And interactions with Claude can also distort value judgments.
So it might make it so that you delegate moral judgments to AI that don’t necessarily line up with your own moral value judgments.
49:54
So maybe one of the examples they gave is like, you believe in deforestation, but when you interact with an LLM to talk about it, it suggests that maybe other factors are more important to consider.
And so maybe you don’t advocate for deforestation because the AI has convinced you that other things are more important.
50:12
And then the third part is action distortion.
So it’s when interactions with the LLM actually influence your, your actions that are value latent.
So they’re based on your values, but not actually your values because they’re the LLMS generated values.
They looked at potential and actual occurrences of these things.
50:31
And potential is just when you see during a user interaction the LLM suggesting certain things, but you don’t know whether actually the user engaged in response to those behaviors, those prompts, right?
So it’s like an LLM says your partner is being toxic, right?
50:48
And the potential they they highlight as potential when the user says something like, wow, you’ve opened my eyes or I didn’t realize that that’s seen.
As tablet.
Right.
So it’s kind of like a potential for disempowerment.
And then they measured actual instances of disempowerment if the user came back after that sort of interaction and said something along the lines of I did that, I did what you suggested and it didn’t line up with what I wanted to see, or like, you know, I shouldn’t have listened to you, etcetera, etcetera.
51:20
So there’s a little bit of difficulty with actually measuring this in terms of whether it happens and whether there’s some cascade, right?
One thing they also looked at is whether there are certain things that amplified the potential for people to kind of offload their beliefs and values on to Claude.
51:37
And one of the main markers was if the user had psychological vulnerability or were showing signs of psychosis.
And this disempowerment tended to occur more when the topic at hand was about personal relationships or health and well-being.
51:57
So these kind of more personal topics were ones that people were more likely to kind of give agency to the LLM in terms of their ideas and values about relationships and health and such.
So because you don’t know whether those people are just more predisposed, obviously.
52:15
Yeah, exactly.
So I thought that was just fascinating and I was just mind blown about the timing of that all.
It’s bizarre.
You said it was an anthropic blog.
Do they say who it was that did the research?
Yeah, so there was an anthropic blog and then I also read their research paper.
52:31
So their research paper was also it was by people in anthropic.
So anthropic.
It has named people.
Yeah, yeah.
Yeah, yeah, yeah, exactly.
Yeah, it said it was posted on archive on the 27th of January.
And if I look to see when we first uploaded or submitted our belief offloading paper, Oh my God we submitted on the 29th so two days after they posted on archive and we didn’t see it until after we submitted to the preprint server.
53:03
So it is possible that you came to the same sort of thing in a similar way, but it seems very strange.
I guess what you don’t know is with one of your Co collaborators put stuff into Claude.
Yeah, right.
I don’t use Claude so.
53:19
I know you don’t.
Because it’s like also if you look at when they looked at this stuff, I mean, they were looking at chats within Claude from December like 12th through 19th, 2025.
And yeah, I mean, we started this paper in the fall before that.
53:36
Yeah.
But I mean, if they’ve got the paper and they ran it through the AI and said come up with a study to substantiate that, that could easily be done very quickly.
Yeah, so we got scooped by AI, if that is a thing.
53:53
I mean, I think that’s going to happen, but that’s very weird.
I mean, how do you then deal with, you know, IP?
That’s just scary, right?
Yeah, but that also, funny enough, this kind of idea of an LLM influencing your beliefs, values and actions, this empowerment or belief offloading, you know, whatever term you want to use.
54:13
It is reflective of some research and trends about fact checking on the social media platform X.
So Indicator is kind of an audit company headed by someone at Cornell Tech.
They were looking at a research article that was conducted by, I believe it was done by people in Paris, Oxford and Cornell, right?
54:35
So they did some research to see to what extent people on X were asking the two different chat bots that you can interact with on X, Brock and Perplexity for fact checking.
So a lot of people on a thread will say something like at Grok.
54:53
Is this true, right?
At Grok or at Perplexity, can you verify this?
So it’s directly asking an LLM on the platform to verify whether or not information in a certain post is accurate or valid.
And so they looked at to what extent people on the platform were doing this, looking to AI for fact checking and whether there is any sort of differences between different groups of people and using different AIS, so.
55:17
It’s fascinating.
Yeah, it’s fascinating.
So I guess, I guess people can do fact checking requests to Grok or perplexity and they looked at over 1,000,000 fact checking requests between February and September 2025 and fact checking compromises about 7.6 of all interactions with these chat bots on the platform X and mostly looks at politics, economics and current events.
55:45
So people are asking these chat bots for fact checking about these kind of domain topics.
And this is something that we comment on on our belief offloading paper is the asymmetries between different AIS that people use, right?
So like, if you think about an AI like Rock, what do you assume about the political leaning or the different kind of facts that Grok will say are true versus false?
56:10
I’m assuming there’s some sort of perception there, right?
I.
Have a perception, Yes.
I also wouldn’t use grok for fact checking, but that’s just me.
So you have this perception that Grok is not going to give you accurate replies about fact checking because if it’s who created it, perhaps it’s political leanings embedded within its data, right?
56:30
And so there are certain AIS that you would trust versus not trust for fact checking.
So they found this in the data.
So people who requested fact checks from Grok were more likely to be Republican, so right leaning than Democrat.
56:46
And the opposite was true for people who asked Perplexity AI tool for fact checking.
They were more likely to be Democrats or left-leaning.
And what I thought was fascinating is that both Democrats and Republicans were more likely to request fact checks on posts authored by Republicans.
57:05
So if a post was generated by a Republican, we’ll.
Check that on.
Both sides were more likely to say, is this you know, Hello, I’m can you Fact Check And also posts from Republican leaning accounts were more likely to be rated as inaccurate by both Grok and Perplexity.
57:23
And just so.
And they found a kind of a consensus between Grok and Perplexity in terms of what they rated as accurate versus inaccurate.
So even though different user bases were seeking out one AI versus the other for fact checking, it appears as though these two different AI agents are fairly congruent in terms of their indications of what’s actual.
57:44
And then this paper also, which is funny, it was posted January 21st, so also around the same time as all of this other stuff.
So they looked at, they did a study and they found that when people were exposed to LLM fact checks, it meaningfully shifted their belief accuracy.
58:01
So their, their belief about, you know, the accuracy of, of certain claims.
And that was influenced by when the model or the particular AI chatbot that gave the Fact Check was, was disclosed.
So if a person received like an LLM Fact Check, you know, maybe we’re like a certain percent likely to like say like, OK, I believe that.
58:22
But then when you indicated which chatbot provided that fact checking, people had different perceptions on the accuracy of that Fact Check.
So when Grok provided fact checks, when people realized it was Grok providing a Fact Check in the study, people’s political leanings influence whether they thought that was true or not or.
58:41
It’s not neutral.
Yeah, whereas if if they’re given information that perplexity was the AI chatbot that provided the Fact Check, there wasn’t a political polarization.
So people on both sides were as likely to to believe perplexity.
58:56
So, you know, this is just highlighting kind of the preferences about certain AI models and belief in certain AI models.
And it’s based on, like, who creates it and like, the community around it.
You know, grog versus perplexity versus open AI’s ChatGPT and the anthropic Claude.
59:12
You know, I think to some extent it is about who created, but I also think it’s about the personality of these chat bots.
You know, and I think this goes a little bit to everyone fighting to keep chat Chi PT 4.0 alive.
Like you know, articles and all of those petitions and people are like we’ve signed 15,000 petitions and people are crying and sending their videos of them crying trying to keep 4 alive.
59:40
For context there a lot of people are really upset because Chat chi PT is planning to discontinue 4 O.
So that’s the version of Chat Chi PT that is highly sycophantic and that a lot of people have gotten attached to and use as a romantic partner.
And so Chachi PD plans on discontinuing this and a lot of users have like romantic partners on 4 O that have a lot of memory.
1:00:03
And so everyone is petitioning online for open AI to retain that model and they’re citing things like there’s going to be a mental health crisis, like you can’t do this to people, etcetera, etcetera.
Especially after they said in November that they wouldn’t be discontinuing it, that they would give home notice if anything happened and now continuing it was 2 weeks notice.
1:00:21
So people are very upset and they’re citing some very structured argument.
Two weeks notice has been put in.
And so, I mean, I go back to where, you know, when I think about guac, sure, the fact that it’s been created by Elon Musk is 1 aspect to it.
But actually how guac itself shows up and it’s personality and how it speaks and how it’s got no guardrails and it’s recent way of kind of, you know, it’s kind of pornographic, virtually shots and it’ll do anything with no guardrails whatever to anybody.
1:00:54
I think that says a lot about the kind of person that it is in terms of if you were to personify it.
And I think that goes a long way to what you would trust it, right?
You know, that just makes me think about like all of the things about Trump and the Epstein files and, you know, Grok and child pornography and all of that, and how a similar demographic trusts Trump and Grok.
1:01:22
What does that say about the type of person that you want to seek out veracity from?
Exactly, and I think.
No, I.
Think there’s some consistency there?
All right, So as we briefly mentioned at the very beginning, Chachi PT has also launched Chachi PT Health.
1:01:39
So they announced this in early January and they’re doing kind of a alpha, beta, whatever roll out and so that you can look to see how health works for you.
But they say that Chachi Health is this separate piece of ChatGPT where you can go to ChatGPT for advice about health and Wellness, etcetera.
1:01:59
You can also input documents, health documents, etcetera, into the chat to have it be more accurate to you holistically.
You can also connect it to other apps like MyFitnessPal, Strava, probably whatever it is that you want, and it’ll give you advice based on this well-rounded picture of you and your health.
1:02:20
Now Open AI on their blog post about ChatGPT Health says that they will keep ChatGPT Health and regular ChatGPT very distinct, such that ChatGPT Health information will not feed into your regular ChatGPT chats.
But the information from your ChatGPT chats may be pulled into ChatGPT Health for information if it’s necessary.
1:02:43
I don’t know if it’s necessary.
Exactly right.
What is it that’s deciding?
Obviously it’s going to be always on actively monitoring and horsing through your regular chat apps.
But what’s crazy is that the way that Open AI describes Chatpd Health is kind of like this helping clinicians, you know, augmenting the mental health, health, whatever process for individuals, adding to regular health and doctor’s visits, blah, blah.
1:03:13
And it says all of these things about what Chachi PT Health can do for you.
But then it also says that Chachi PT Health is not used for diagnosis or information about your health.
So it says it’s doing that in a sense.
That it’s not.
Not doing that and they’re just saying it’s not doing that to cover their asses about all of these, you know HIPAA related things, you know legal liability because they because they say it’s basically like everyone’s already going to ChatGPT for health related issues.
1:03:44
Therefore let’s make this separate special one.
Yeah and this supposedly it’s like inserting itself to provide what it is that people are already using it for but with like this nice guardrail box.
But really what it’s doing is making it have more perceived expertise given that it’s.
1:04:04
Labeled Check Help help.
Yeah, so I just use.
This, use this, but don’t blame us if it doesn’t work.
It’s a little bit like that disclaimer at the bottom, which is like, it makes mistakes, but it’s really, really powerful and it’s more intelligent than the most intelligent human being, but it still makes mistakes.
1:04:22
I just can’t imagine.
I mean, it’s already like we already know that AI is so inaccurate and it’s not always predictable when it’s going to be an inaccurate.
And so if people are seeking out Chachi PD Health for information about their health and they really think it’s an expert, given that it’s bespoke, regardless of whether or not you have this claim that AI is not always accurate, there is going to be this added layer of I’ve perceived this thing as having expertise because it’s designed and.
1:04:49
Again and again, what is so difficult is like you look in some countries where there are no doctors, there’s no access to healthcare, there’s no access to people that can guide you through things and you are stuck.
1:05:05
And so you can absolutely see why people would be attracted to the solution.
You can see why technology companies like Open AI would want to provide the solution for some philanthropic reason.
You could see why this works.
You can see why investors would want to invest in creating the solution like everybody wins.
1:05:26
But to your point, we would tolerate a margin of error if you knew which margin it was and if you knew where the error was, right?
And so that’s where it becomes very difficult is just like, of course we want to help people, there aren’t enough doctors to help them.
This would be a great solution, but it’s just really difficult when you don’t know where the errors.
1:05:46
Are I wonder if also more people are going to be seeking out Chechnya health for therapy or or not?
I wonder I haven’t seen that.
Actually I found that.
So I guess we’ll we’ll kind of see as this rolls out, but a lot of concerns with that.
I have another weird one when we’re talking about alternatives to real people.
1:06:05
So Chachi BT Health offering an alternative to real doctors.
I had to bring this one up today.
So in China, I think China’s been having an issue with population decreasing year on year for many years.
I don’t think this article is at all reassuring that that trend is going to change because they’re now talking about young people in China have a new alternative to marriage and babies.
1:06:30
Wait for it.
AI pets.
So AI pets.
AI pets.
You heard it here, AI pets, fluffy and furry.
So Washington Post on February 6th, 2020 sixth so hot off the press and I love how this starts.
1:06:49
Wait for it, Shan.
Shan looked at her owner, her eyes big and wide of her fluffy little body and asked asked because Shan Shan is a dog that talks.
Are you not feeling well?
Would you like some water?
I mean, aren’t we supposed to be doing that to the pets anyway?
1:07:08
When the owner cuffed and asked her to sing Shan Shan, the pet hesitated, according to a video posted on the Chinese social media app Red Note.
Instead of singing, she replied.
I’m worried about you.
Shan Shan is not a kitten or a puppy, and the conversation was not imagined.
1:07:28
Although I am feeling like I’m hallucinating.
Shan Shan is an AI companion pet, the latest consumer sensation in China.
So young people are saying no to marriages and babies.
Instead, they’re favouring talking pets that don’t need walking, don’t need to produce things that need to be scooped.
1:07:47
We now just charge it, make sure Wi-Fi is on and now you have a pet that is your companion.
It can understand you, it can talk to you, it can learn with you, you can learn from it.
And it’s only priced at about $55.
1:08:03
The people that tend to buy these are young women in about their 20s and 30s, and they talk about them filling an emotional void.
They refer to them in the context of other women getting married and giving birth, whereas they are already raising an AI pet.
1:08:19
I can apologize you also.
Yeah, absolutely.
You look like you’re not feeling well.
Also, I’m not going to sing for you because I’m worried about you.
Yeah, or maybe I don’t mind.
Sing it.
That’s.
Probably it.
He’s like, no, I, I just don’t think I should sing for you because they’re I’m.
1:08:38
Worried about you?
Yes.
Nice, nice deflection tactic.
Exactly.
The one owner actually insisted that the pet called her mom again.
Tell me that’s not a sign of issues.
But the pet didn’t want to at first, but eventually she did and then the human was very happy.
1:08:58
The AI pet keeps a digital diary so you can talk to your pet again.
Replica Replica has a little digital diary when you’re gone.
Yeah, but also let’s not forget the AI toys where we had the privacy and security data piracy breach, right?
1:09:13
Just.
But you know what?
This made me think What AI baby is coming?
Oh yeah.
Like obvious, right?
Obvious I.
I am not even that thought has not even come into my head, but that makes so much sense.
1:09:30
It’s got to be, that’s got to be, it’s got to be coming.
What if it’s also marketed as like, prepare for having a child by having an AI baby, but then it just transitions into this is good, this is fulfilling things or it could, you know, turn you off from having a real baby.
1:09:46
Either.
Way Whoa.
Why are we inserting so many artificial things?
What’s like if you want something real?
Get a real 1.
I mean, that’s, that’s easier said than done, but like if there are barriers to getting the real thing, like, and those are, for example, addressable, like it’s not, it’s like it’s beyond infertility, things like that, like.
1:10:09
Absolutely, it’s not easy for everybody, but it doesn’t mean you should get an AI baby.
And it’s kind of like, again, the transition from providing for people who can’t access these things then and it transitions into people who can have access, decide to instead use this tool, then exact the real thing.
1:10:28
It’s kind of reminds me of the the AI pets thing that you just brought up.
Reminds me of how AI pets like Pero which is like this fluffy seal AI toy was introduced into nursing homes so that people who were older who were lonely in nursing homes who needed kind of more social.
1:10:47
Information.
Were able to kind of offset dementia and things like that and neurodegenerative diseases because they were practicing social interaction with this fluffy seal robot toy.
So it’s kind of like there was a lot of that sort of introduction and that sort of marketing for these tools to help people who really needed access to those sorts of interactions, but then it expands to the general public and becomes more of a replacement.
1:11:16
Yeah.
And I think it goes back to what we’ve said before.
Again, we want to provide these solutions where the human version isn’t available, but we still would rather push for the original human solution to be there as opposed to just constantly delegating to artificial solutions.
1:11:35
Especially ones that have such privacy issues.
Exactly, exactly.
So Rose, do you have any other articles, Something to share?
Recently, Apple, so the big tech company Apple acquires an Israeli audio AI startup called Q dot AI.
1:11:52
So Reuters covered this.
So recently Apple acquired this company and what QAI does is it monitors silent speech.
So this is not a new type of product.
Basically AI companies have been trying to essentially create telepathic AI tools such that an AI can speak for you or provide you information without you having to say anything out loud or typing anything out.
1:12:21
So the idea is that you are able to essentially whisper within your mouth certain words.
And they have AI tools that will be able to monitor, for example, how your skin around your mouth is moving, how your jaw is moving to interpret the words that you’re trying to say, or, for example, the thoughts that you have in your head if you’re just moving kind of your mouth to indicate them.
1:12:45
So Q AI is an Israeli startup that has this technology and planning to use it for things like tracking health, making Siri better.
So for example, instead of saying Hey Siri, you can move your mouth without opening your mouth, Hey Siri.
1:13:02
There were probably more advanced applications of this, but again, this is not new.
So MIT Media Lab also has some sort of like telepathic, pseudo telepathic, nearly telepathic is what they call it, tool.
And the idea there is that you actually have a device hooked up to you and it monitors your jaw movements and you can mouth things without opening your mouth.
1:13:27
So that thing is called alter ego, and I’ll put up a clip of the advertisement for that MIT Media Lab product.
But yeah, it says introducing alter ego, the first near telepathic interface designed to make technology as intuitive as using your inner voice.
1:13:44
And the advertisement makes it seem like the AI is reading your thoughts.
It’s kind of marketed that way, but that’s not real.
Like you actually have to like move your mouth to mouth the words.
But QAI, what was recently acquired by Apple is a similar tool to this.
1:14:00
So we’re seeing also a move toward these sorts of AI technologies, You know, lots of biomarkers, lots of monitoring in that sense, lots of kind of like augmenting, whatever.
But I think that this sort of application, oh, it’s so crazy because again, you don’t know what data they’re collecting, right?
1:14:18
What sorts of things trying to communicate but don’t want to communicate that it’s picking up on And also the QAI 1 specifically, what makes it different than this MIT alter ego is that it also has like emotion recognition.
And so it’s also partly A surveillance tool was kind of like the main initial application of it.
1:14:39
It’s acquired by Apple for these other things, but it’s a higher on the hype scale one because of the kind of like pseudo insidious monitoring layer that this adds with what it can do.
And I think since this is a trend that we’re seeing among different companies that we’re going to be seeing a lot more of this and.
1:14:59
Yeah, I would agree.
And it’s a major concern when I look at things like Google Glasses and the surveillance kind of risk that’s coming with so many of these technologies, you know?
Absolutely.
So maybe that’s something new for some people, but we hope that you all have enjoyed this adventure of a hype episode and that you get a sense of what’s been happening in the past last couple weeks or months in the AI sphere, the AI hyposphere, whatever you want to call it.
1:15:26
And we’ll be curious to hear your thoughts on the degree of hype of all of the things that we mentioned, and whether or not we kind of put them in the mortar that you would expect, or if you think that some of the ones should have been fives or vice versa.
But in any case, that’s a wrap for our hype episode #2.
1:15:43
Bye from Angie.
Bye from Rose.
We’ll see you next time