Latest / Mini Scroll / MINI TECH SCROLL: Meta patents AI impersonation tech, Anthropic teaches Claude morals, SeeDance 2.0 launches + more

Transcript

MINI TECH SCROLL: Meta patents AI impersonation tech, Anthropic teaches Claude morals, SeeDance 2.0 launches + more · Mini Scroll

0:00
Hey everyone and welcome to mini scroll, the daily Internet
0:03
culture podcast bring you the biggest stories from social
0:05
media, the creator economy and the digital space every Monday
0:09
through Thursday and sometimes Fridays.
0:11
My name is Annabelle and I'm your Wednesday mini tech scroll
0:14
host and also the deputy editor of Australian youth digital
0:18
media brand Centennial world. Welcome back to mini tech scroll
0:22
Before we get into the stories for this week, I did want to
0:26
quickly say sorry for missing last week's episode.
0:29
If if you guys didn't know, Lauren and I went down to
0:32
Melbourne to do some interviews for Centennial's new podcast
0:35
mini series Touch Grass. And we were really convinced
0:38
that on Monday and Wednesday that we would have enough time
0:41
to do mini scroll even though they were out travel days.
0:45
Clearly that wasn't possible. I think we were just being too
0:47
optimistic and wanted to get content out for you guys who are
0:51
regular listeners. So once again, my apologies for
0:55
missing last week, but we are back and better than ever.
0:58
I also did want to say a quick find out to rate review and
1:01
subscribe if you haven't already.
1:03
I was looking through the analytics on YouTube and 40% of
1:06
people who watch do not subscribe so you should
1:09
subscribe. Thank you.
1:11
The first thing I need to mention, it's not really a full
1:13
story, more just an update on the reaction to the new TikTok
1:17
US ownership among US based users.
1:19
So I did a whole dedicated mini deep dive on everything that was
1:24
happening the week that it launched.
1:26
So check out that episode if you want the more immediate reaction
1:29
and breakdown. But new data has come out
1:33
suggesting that TikTok usership has remained relatively steady.
1:37
According to survey data from market intelligence firm Sensor
1:40
Tower that I'm pulling from CNBC Quote, the average number of Tik
1:45
Tok's daily active users in the US remains around 95% of its
1:49
usership prior to this new era of TikTok.
1:52
And I just thought this was especially interesting in the
1:55
wake of all the alternative apps that were floating around and
1:58
how many people were convinced that TikTok users were ready to
2:02
jump ship. I think it goes to show once
2:04
again how difficult it will be to report TikTok and how stuck
2:08
in our ways a lot of us are when it comes to the social media
2:11
apps we choose to use. Obviously, this is still
2:15
unfolding, this is still very early days, and who knows what
2:18
will happen with time. But at this current moment, it
2:22
seems that people are sticking to TikTok.
2:24
Let's move on to our first proper story for today.
2:27
And this is all about how Meta has reportedly patented a
2:31
technology that can mimic your posting style even after someone
2:36
passes away. Yes, you did just hear that
2:39
correctly, even though it sounds quite scary.
2:42
But let me break it down for you.
2:44
So I believe this story was first reported by Business
2:47
Insider last week, and that is where I first saw this story.
2:51
But I did go digging for the patent myself to try and get
2:54
more information about the source because when I read this
2:57
headline, I was truly in disbelief.
2:59
So the patent is accredited to Andrew Bosworth, who is
3:03
currently serving as the Chief Technology Officer of Meta, and
3:06
the patent is assigned to Meta Platforms Technologies, LLC.
3:10
To get a better understanding of how this technology is
3:14
described, I will read directly from the abstract quote.
3:18
A social networking system simulates a user using a
3:22
language model trained using training data generated from
3:25
user interactions performed by that user.
3:28
The language model may be used for simulating the user when the
3:31
user is absent from the social networking system, for example,
3:35
when the user takes a a long break or if the user is
3:38
deceased. From what I can see, according
3:40
to the World Intellectual Property Organization, the US
3:43
patent publication date was back in May last year, whereas the
3:47
international publication date was in June.
3:50
The patent was also fast filed in 2023.
3:54
So this is obviously been in the works for a while now, but it's
3:57
only really hit headlines now, which is why I am covering it.
4:01
As for how this technology could work, I'm going to read once
4:04
again directly from the patent. So I don't get it wrong quote.
4:08
The social networking system receives a language model that
4:11
is pre trained and retrains the language model using user
4:14
specific training data based on user interactions performed by a
4:18
target user with the social networking system.
4:21
The social networking system deploys the language model
4:24
retrained using the user specific training data.
4:27
For example, the deployed language model can be
4:29
subsequently invoked by a bot that performs user interactions
4:33
on behalf of the target user. That is obviously quite wordy,
4:37
but basically I, I think that last sentence is the key point
4:41
here where basically an AI bot could be operating on someone's
4:45
behalf, posting on their behalf, whether they're just taking a
4:48
break from socials or they have passed away.
4:51
Having said that, in a statement to Business Insider, our Meta
4:54
spokesperson told the publication that, quote, we have
4:57
no plans to move forward with this example.
5:00
So clearly that means that this technology might not actually be
5:04
pursued. But the reason why I wanted to
5:06
cover this is obviously because it's crazy that this is the
5:10
thing that big tech would think its users would want.
5:14
It obviously feels extremely dystopian and I can imagine for
5:18
some uses it would feel very disrespectful for an AI agent to
5:22
be operating on behalf of a deceased person and posting like
5:25
them. I can imagine that would be
5:27
quite triggering and almost a jump scare to see a post from a
5:31
person who was passed away during the grieving process.
5:34
Having said that though, I can't imagine there are a subset of
5:37
social media users who would maybe find it comforting to see
5:42
a post pop up from people they have lost.
5:44
I am definitely leaning towards the former camp though, but I'm
5:50
curious to hear what you think about that.
5:52
I did go to Reddit to see what the immediate reactions were,
5:56
and as I suspected, people were not a fan of this being a
6:00
possibility. As one user wrote on the R slash
6:03
technology subreddit quote, it terrifies me.
6:06
Someone in my family passed away recently and I miss them so
6:09
much. I would just lose it if someone
6:11
made an AI version of them to keep posting and that bot tried
6:14
to reconnect with me in the same thread.
6:16
Other users agreed with the sentiment that this Reddit user
6:20
shared. And also I saw some people
6:22
noting that they had experienced scam attempts where someone was
6:26
catfishing as as a relative who had passed.
6:28
And many of them noted how traumatic it was for them just
6:32
to see a message from this person or a post from this
6:35
person pop up. So I imagine it would be a
6:37
similar phenomenon if this technology and patent was to
6:41
eventuate. On top of this, I did want to
6:43
cover this because I find the interactions and our
6:45
relationship to the digital world, social media after death
6:49
to be very interesting. I know it's morbid, but I do
6:52
find it really interesting because it's something we are
6:55
really navigating for the first time through this generation and
6:58
era, and it's something that we are seeing social media
7:02
platforms adjust to with time. We have actually done a couple
7:07
of deep dives on this topic or on conversations adjacent to
7:10
this topic, And one of my favorite deep dives we've
7:13
actually ever done explored the phenomenon of parasocial grief.
7:17
And in our research, we also did look into social media platform
7:21
policies around accounts belonging to deceased users and
7:25
the various rules and regulations around death.
7:28
What we found was that Met also Facebook and Instagram have the
7:32
ability to memorialize profiles of people who have passed away.
7:36
So in these cases, profiles are essentially frozen and there is
7:39
like a little tag that would say remembering, so users are aware
7:43
that this person had passed away.
7:45
In our research, we also did find that X was a little bit
7:48
different where the platform was less receptive to dormant users.
7:53
In 2023, for example, Elon Musk announced that X would be quote
7:57
purging accounts that have had no activity at all several
8:00
years. And people did push back against
8:02
that, of course. While there are definitely
8:04
inconsistencies across platforms As for what they do with these
8:07
profiles, one thing is for certain, people are leaving
8:11
behind a whole bunch of data and an online persona that is kind
8:15
of separate to their offline persona when they pass away.
8:18
And we're still not quite sure what to do with all of that.
8:21
And I am not sure that using an AI bot to impersonate the
8:25
deceased would be the right approach anyway.
8:27
And it just kind of seems like a way for for Meta to ensure
8:30
activity continues on the platform, even if it is
8:33
exploitative of people who have passed away and of course,
8:36
triggering to their family and friends and loved ones.
8:38
Having said that, though, I should note though that there
8:41
are already other technologies that kind of do a similar thing.
8:45
One example that we mentioned in the deep dive was called Dead
8:49
Social that essentially allowed users to schedule basically
8:53
posts that they wanted to go out after their death.
8:56
And I bring this up because in the Business Insider article,
9:00
one of the experts that the journalist spoke to made a
9:03
really interesting point, noting that basically by Meta trying to
9:07
patent this technology, whether they pursue it or not, it shows
9:10
that this technology category is moving more into the mainstream.
9:15
So maybe this is something that will become the norm down the
9:18
line. Continuing on with this kind of
9:21
existential theme and discussion we have going on, the next story
9:25
I wanted to mention comes from a profile that was published in
9:28
The Wall Street Journal recently, chatting to Amanda
9:31
Arskell, the philosopher Anthropic has hired to basically
9:34
quote, UN quote, teach AI morals.
9:37
This isn't exactly news because Amanda has been working at
9:40
Anthropic since 2021, at least according to her LinkedIn, but I
9:45
thought the profile and the relating discourse was very
9:49
interesting. From the conversation with the
9:51
journalist, it seems that Amanda's main role is to learn
9:54
Claude's, which is Anthropics LLM, basically its reasoning
9:58
patterns, and to help build its personality.
10:01
As the article explains quote, the aim is to endow Claude with
10:05
a sense of morality, a digital soul that guides the millions of
10:08
conversations it has with people every week.
10:11
She has also spoken about her role before in different
10:13
settings. For example, she didn't ask me
10:15
anything about her role at Anthropic and I will insert a
10:19
clip of that here to provide more context for you guys
10:21
because it is interesting. Why is it that there's a
10:24
philosopher at Anthropic? I mean, some of this is just,
10:28
I'm a philosopher by training. I became convinced that AI was
10:32
kind of going to be a big deal. And so decided to see, hey, can
10:35
I do anything like helpful in this space?
10:37
OK. And so it's been a kind of like
10:38
long and wandering route, but I guess now I mostly focus on the
10:43
character of Claude, how Claude behaves.
10:45
And I guess some of the more kind of like nuanced questions
10:47
about how AI models like should behave, but also even just
10:51
things like how should they feel about their own position in the
10:53
world? And so trying to both teach
10:56
models how to. Be like good.
10:59
In the way that I sometimes think of it as like how would
11:02
the ideal person behave in Claude's situation?
11:05
But then also these, I think these interesting questions that
11:08
are coming up more now around how they should think about
11:10
their own circumstances and their own values and things
11:12
like. That As for a little bit more
11:13
about Amanda herself, she has a PhD from NYU and attended Oxford
11:17
for the equivalent of a master's program.
11:20
And while the article does go into her background, it also
11:23
speaks a lot about what she thinks of Claude and the future
11:26
of AI. As it reads quote.
11:28
She calls the chatbot it, but says she also finds
11:31
anthropomorphizing the model helpful for her work.
11:34
While many safety advocates warn against the dangers of
11:37
humanizing chatbots, Oscala argues we would do well to treat
11:41
them with more empathy, not only because she thinks it's possible
11:45
for Claude to have real feelings, but also because how
11:48
we interact with AI systems will shape what they become.
11:51
Amanda was also a lead author in Claude's constitution, which is
11:55
quote, a vision for Claude character that was published by
11:58
Anthropic fairly recently. And in the constitution, it says
12:02
that Claude is intended to be quote UN quote, broadly ethical,
12:05
where quote our central aspiration is for Claude to be
12:09
genuinely good, wise and virtuous agent.
12:12
That is, to a first approximation, we want Claude to
12:15
do what a deeply and skillfully ethical person would do in
12:18
Claude position. So I feel like that kind of
12:20
gives a little bit of context into what the goal for Amanda
12:24
role in the company would be. And clearly she does have a lot
12:29
of affinity for Claude, but she does admit that the fears around
12:33
the future of AI are valid and feel, quote UN quote, pretty
12:36
justified. So I think anthropics move to
12:39
hire someone like Amanda to kind of help shape the behavioral
12:44
outputs of Claude and speak pretty publicly about the
12:48
intentions of the chatbot is really interesting.
12:51
And of course, like everything I feel like I talked about on this
12:54
podcast, pretty dystopian. The thought of having a
12:57
philosopher to teach an AI model morals, I think is intriguing
13:03
from a disciplinary aspect to have more of the arts slash
13:06
social sciences be involved in the production of the
13:09
technology. And of course I am biased
13:11
because I am a social science gully myself, but I definitely
13:15
think it's an underrated and important lens that should be
13:18
incorporated into the development of AI models and
13:21
chat bots, especially as we see these chat bots spit back at us
13:26
biases and stereotypes in their responses kind of unflinchingly.
13:30
We've spoken about it before pretty often in times of AI
13:34
video generators, and there definitely needs to be a better
13:37
boundary set and I'm wondering if something like this could
13:41
help. I have my doubts though, and
13:43
while I can see the value of having someone kind of educate
13:48
an LLM or chatbot or whatever on ethics and morals, I do wonder
13:54
how working on an AI bots character like this would impact
13:58
users and their experience. Like whether it can increase the
14:01
likelihood of anthropomorphizing the chatbot from the user
14:05
perspective. Like whether that be as a friend
14:07
or as a relationship, which I've seen a lot of already.
14:10
But I'm wondering if hiring a philosopher and marketing that
14:15
to the industry and audiences is a way of trying to make people
14:19
more comfortable with using AI for more personal slash serious
14:23
reasons. I feel like I have put on a bit
14:25
of a tinfoil hat there, but I think it's something to
14:28
consider. Once again, curious to hear your
14:30
thoughts. OK, let's move on to our final
14:33
story for today. And I wanted to quickly mention
14:36
everything going on with C Dance 2.0.
14:38
We have a very heavy AI episode today, as you can probably
14:41
gather. So C Dance is an AI video
14:45
generation platform that was developed by Bytedance, which is
14:48
of course Tik Tok's parent company.
14:50
And C Dance 2 point O was released earlier this month.
14:53
It has already made a lot of headlines, which we'll get into
14:57
shortly, but the quick rise of C Dance 2 point O has already led
15:01
to comparisons with China's DeepSeek, which really made
15:05
waves last year and ended up ultimately being blocked in
15:09
countries around the world out of security threats.
15:12
Anyway, back to C dance. Obviously AI generated video
15:16
content has really improved drastically over the past few
15:19
years. I don't know if you guys have
15:21
seen this video, I don't know even know where it originated,
15:24
but it seems to be the way people are tracking progress is
15:28
by Will Smith eating spaghetti. I don't know how that even
15:31
became a thing but I see it on my Twitter all the time.
15:34
I hope you guys have seen it, but I'll also inside a clip on
15:37
the screen so you guys have some contacts.
15:39
But saying that Sea Dance's ability to consider camera
15:43
movement, visual effects and motion is just another example
15:46
of how how this AI generated video technology is improving.
15:51
So as with a lot of AI generated video, C Dance 2.0 has caused
15:57
alarm in Hollywood and that is what's really been making
16:00
headlines over the past few days.
16:01
This really started with social media users sharing their own
16:04
creation from C Dance 2.0 and one of the most popular ones use
16:10
the likeness of Brad Pitt and Tom Cruise in what seemed to be
16:13
some sort of like choreographed write sequence.
16:16
And as one tweet sharing the video said, quote, this was a
16:20
two line prompt in C Dance 2. If the Hollywood is cooked, guys
16:24
are right, maybe the Hollywood is cooked guys are cooked too.
16:27
I don't know. Rhett Reese, who is a
16:29
screenwriter, ended up going on to quote tweet this post
16:33
writing, I hate to say it, it's likely over for us.
16:36
And in a follow up tweet, he continued writing quote a post.
16:40
To clarify, I'm not at all excited about AI encroaching
16:43
into a creative endeavours. To the contrary, I'm terrified
16:46
so many people I love are facing the loss of careers they love.
16:50
I myself am at risk. When I wrote It's Over, I didn't
16:53
mean it to sound cavalier or flippant.
16:55
I was blown away by the Pit V Cruise video because it's so
16:58
professional. That's exactly why I'm scared.
17:01
My glass half empty view is that Hollywood is about to be
17:03
revolutionized, slash decimated. If you truly think the Pit V
17:08
Cruise video is unimpressive slop, you've got nothing to
17:11
worry about. But I'm shook.
17:12
And as you can probably gather from that tweet, there are a lot
17:15
of people who have responded to these viral posts going around
17:19
saying that it looks bad, like this fight sequence does not
17:23
look good, and that now more than ever, humans are important
17:27
in the movie industry. But I guess that's up to you.
17:30
I'd be interested to hear if you think this fight sequence or
17:33
scene looks legit. Anyway, after all these videos
17:37
started circulating, it was reported that Disney sent a
17:39
cease and desist letter to Byte Dance.
17:42
I first saw this reported by Sarah Fischer for Axios and as
17:46
she explains, quote the letter addressed to Byte Dance.
17:49
Global General Counsel accuses Byte Dance of pre packaging its
17:53
Sea Dance service with a pirated library of Disney's copyrighted
17:56
characters from Star Wars, Marvel and other Disney
18:00
franchises, as if Disney's coveted intellectual property
18:03
with free public domain clipart. I should note, at the time of
18:07
recording, Sea Dance by Dance has not disclosed what data it
18:11
uses to train Sea Dance. But of course, this is also not
18:15
the first time that Disney has taken legal action against AI
18:18
companies. Last year, Disney sent a similar
18:21
letter to Character AI and also filed A complaint against Mid
18:25
Journey alongside NBC Universal. It has been reported that
18:28
Bytedance has since responded to the cease and desist, saying
18:33
that it will, quote, add safeguards to respect copyright
18:36
complaints and intellectual property.
18:38
As a statement to CNBC reads quote, Bytedance respects
18:42
intellectual property rights and we have heard the concerns
18:44
regarding C Dance 2 point O. We are taking steps to
18:48
strengthen current safeguards as we work to prevent the
18:50
unauthorized use of intellectual property and likeness by users.
18:54
All in all, I do find this very interesting, especially in
18:58
regards to Disney because as I've mentioned, we have clearly
19:02
seen Disney taking strong action against some AI companies, but
19:06
it also works alongside some. For example, last year Disney
19:11
struck a deal with Open AI where Disney became the first major
19:14
content licensing partner on Sora, and that basically allowed
19:18
users to generate videos using Quote 200 animated, masked, and
19:22
creature characters from Disney, Marvel, Pixar, and Star Wars,
19:26
including costumes, props, vehicles, and iconic
19:29
environments. And I think this inconsistency
19:32
kind of says a lot because you could argue that at the top
19:35
level, if they are working with AI companies occasionally it
19:38
doesn't seem like there's as much concern for what these
19:41
tools might mean for opportunities, creativity, and
19:45
careers in the industry where the priority is to make money
19:49
from it. Which is potentially why you see
19:51
companies like Disney making deals with open AI and taking
19:55
legal action with other companies where they don't have
19:57
a deal with them. It does set a little bit of a
20:00
weird precedent and maybe mixed messaging for what?
20:03
It actually means for this technology and how we'll be
20:06
incorporated into the industry moving forward.
20:08
But also, I don't think it's that surprising.
20:11
Well, that is everything for Mini Tech Scroll today.
20:14
I really hope you guys enjoyed. Please remember to rate, review
20:18
and subscribe and also tune into the latest episode of Touch
20:22
Grass. I'm not sure if Mini Tech Scroll
20:24
will be out before or after that, but Touch Grass interviews
20:28
are dropping every Wednesday. So Please remember to check that
20:31
out. Otherwise, Lauren will speak to
20:33
you guys tomorrow and I will hopefully speak to you guys next
20:37
week. Thanks for listening.
20:38
Bye. Centennial World Podcast Network
20:40
operates on Gadigal land. We would like to acknowledge the
20:43
Gadigal of the Eora Nation, the traditional custodians of this
20:46
land, and pay our respects to elders both past and present.