July 10, 2026 • 44 mins
Do chatbots have feelings? Anthropic, Google, and Meta are hiring computer scientists, neuroscientists, and philosophers to find out. Nitasha Tiku (The Washington Post) explains why the search for AI consciousness has gone from fringe theory to Silicon Valley mainstream, and why the answer matters more than you think. Then, Taylor Lorenz (User Mag) traveled cross-country to attend an anti-big tech festival in New York City called the Summer of Ludd. She tells us about cleansing her laptop, talking to the event’s puppet spokesperson, and learning what these young organizers hope the future holds. Finally, Reed Albergotti (Semaphor) on what happens when your AI coding agent goes rogue while you're on vacation in the Alps — and the heroic lengths required to reach an actual human at OpenAI.
Additional Reading:
- AI Makes Mistakes, Too | Semafor
- They built the world's most powerful AI. They're facing a mystery they can't explain. | The Washington Post
See omnystudio.com/listener for privacy information.
Listen
Watch
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:11):
So you've all likely heard about Tilly Norwood, the first
ai A lister. Her creator just announced that Tilly is
about to star in her first feature film. Taylor and Natasha,
your face is a quite a picture.
Speaker 2 (00:26):
This is just a fake PR thing.
Speaker 3 (00:28):
Wait, Taylor, do you remember when all the editors wanted
you to like interview Little Michayla.
Speaker 1 (00:34):
Yeah, I've met Lit Michayla's creator, Trevor.
Speaker 2 (00:39):
I know Trevor.
Speaker 4 (00:40):
I've known Trevor for a long time. I have the
girl that Little Michaela was designed off of. I But
with all due respect, love Trevor. But guys, this is
just PR.
Speaker 2 (00:49):
It's just PR.
Speaker 1 (00:54):
Welcome to tech stuff. I'm Maza Velosa and this is
the Weekend Tech, where I'm joined by the world's most
plugged in report is to break down what's really happening
in tech right now. Today, we're joined by Taylor Lorenz
of User mag Read Abrigotti, tech editor for Semiphore, and
Natasha Tku, tech reporter at the Washington Post.
Speaker 5 (01:11):
Welcome, Well, how's it going.
Speaker 2 (01:12):
Thanks for having us.
Speaker 3 (01:13):
Nice to be back.
Speaker 1 (01:15):
Natasha. You recently published an article that got to number one,
I think in the in the Washington post algorithm. The
subheads of the story is anthropic. Google and Meta have
hired computer scientists, neuroscientists, and philosophers to study what some
in the industry think may become a moral crisis. The
story is about AI consciousness.
Speaker 5 (01:36):
What is that?
Speaker 3 (01:40):
Oh, start with the hard one. Huh. You know, it's
this idea that has percolated in the back of the
AI industry, like on the fringes for decades. You know,
the same people who hope to upload their consciousness one
day or you know, become transhuman and merge with machine
(02:00):
also anticipated this idea that you know, sufficiently advanced AI
would have subjective experiences and goals and preferences, just like
human beings. And you know, I first got interested in
this subject four years ago, Like it was a few
months before chat chipet came out, and I wrote a
(02:20):
story about this Google engineer who believed the company's large
language model was sentient, was conscious in some way, and
he got fired.
Speaker 1 (02:29):
Right, he was a whistleblower or is that right?
Speaker 3 (02:32):
I think that's fair to say, Yeah, I'm.
Speaker 1 (02:34):
Not a whistleblower per se. He viewed himself as a whistleblower.
Others viewed him as crazy.
Speaker 3 (02:38):
I think that's also fair to say. Yeah, so he
you know, at the time, none of us had like
it wasn't public, You couldn't like talk to these very
human like machines. So I thought, you know, it was
a kind of response to this technology designed to sound
like a human and you know, often used first person
and refer to its own experiences. And I think after
(03:01):
that the tech companies stepped back a little bit. You know,
they wanted people to adopt their very productive tool. They
were showing all the practical use cases. And I would
say over the last year or so for Anthropic and
then the last few months for other companies, Google, Meta
and Anthropic have really embraced this idea in public. I mean,
(03:23):
we've talked on the show before about Chris Ola, the
Anthropic co founder, talking about, you know, can chat pots
have experience emotions and you know, joy and despair on
stage at the Vatican a few seats away from the Pope, and.
Speaker 1 (03:38):
As you point out your story that was a point
of difference between him and the Pope because the Pope
expelicitically said no, right.
Speaker 3 (03:43):
Yeah, exactly, Like it's still unclear who was who got
the upper hand there. And who was using who you
know to push their their message. But yeah, I mean
they really just don't shy away from it now. Like
Anthropic talks about its AI psychiatry team. You know, they
hired this guy, Kyle Fish, who actually used to work
(04:03):
in like bioengineering for meat replacement, but now he's their
head of model welfare and thinking about like changes that
the company could potentially make to ensure that if they
find out that the models are conscious, they're meeting their
like moral obligation. It just very quickly goes to analogies
(04:27):
around slavery and genocide or factory farming. So so you know,
and from the mouth of these executives at the tech companies.
Speaker 6 (04:37):
So I just he's still in the meat replacement industry.
Apparently I just thrown a dad joke.
Speaker 1 (04:47):
Farewell to us some meat puppets read tailor I mean,
what do you do you think this is on behalf
of the technology companies are kind of legitimate and even
necessary line of inquiry or do you think this is
kind of you know, or a boros kind of snaky
in its own tail of falling in love without creations?
Speaker 5 (05:05):
Yet again, I.
Speaker 4 (05:06):
Think it's deranged. I don't think there's any world in
which an LM is conscious. It reminds me of that
rain Fisher Kwandswait after a judge of et came out
where she typed, you know, typing I'm evil in a
Google doc and getting scared. Basically, I think this is
very silly, and I think it's going to blow up
in their faces. I think this is completely going to
(05:28):
give consumers the wrong idea. The more that people consider
this thing sentient, the more they're going to sort of
basically expect it to have a morality and a moral compass,
and it's never going to be able to do that.
Speaker 6 (05:40):
Yeah, these these companies have now, you know, put a
lot of guardrails around these things so that they don't
actually sort of seem as human like, as sentient as
as they probably would if you use these things with
like you know, with the raw model, they sort of
they're a lot more human like and a lot crazier honestly.
(06:00):
So they I think it's it's like, I think, hard
for us to understand like what people like the guy
you wrote about four years ago we're seeing, I think,
But it is ultimately just you know, text prediction, right,
There's nothing sentient about it. I think that I think
these companies are I mean a lot of these people
inside them are. They're very genuine people and they do
(06:22):
have these concerns, but they're also like, in a way,
it's marketing because it's like these things are going to
end up being so intelligent, they're going to get to
human like intelligent, what do we do? What do we
do when we get there? And like we may never
get there, right, it's not ultimately it may not ultimately
be possible to even build what they're sort of concerned about.
(06:44):
I think that's another way of looking at it.
Speaker 1 (06:46):
Natasha, you write a real piece at the search for
machine consciousness is practically mainstream. Do you mean mainstream within
the type of technology companies or yes.
Speaker 3 (06:55):
Yeah, yeah, I mean but the fact that it's happening
in peril well, with so many people asking whether or not,
you know, the chatbot that they're talking to every day
is alive in some way or human in some way,
I just find that remarkable. I mean, I think you
know what Taylor said about it, coming back to bite
(07:16):
the companies is super interesting because I wanted to write
about this because it is a prelude to an argument
about personhood and rights for the AI. So I think
actually the companies are hoping that they won't be accountable
for some of the actions of their chatbots in the future.
You know, they're really like, this is only this argument
(07:38):
is only going to get weirder.
Speaker 1 (07:39):
But they want to push the liability onto the chatbots.
I mean you would you would think, I mean, I
would think anyway that it would be bad for them
for the chatbots to be considered conscious or having any
kind of self of because that would open them up
to a different type of liability.
Speaker 3 (07:54):
Right, Yes, unless they're able to confer personhood on a chatbot,
in which case you can't. Yeah, you can't go after
the anthropic engineer.
Speaker 5 (08:05):
Has anyone actually told you that?
Speaker 3 (08:07):
Has anyone told me that about personhood?
Speaker 2 (08:09):
Yes?
Speaker 6 (08:09):
Has anyone at these companies said their strategy is like,
this is going to be our way out of liability.
Speaker 3 (08:14):
No, no, no, no, I'm going to give them more liability.
Speaker 2 (08:17):
They're going to have more liability.
Speaker 5 (08:19):
If anybody thinks that they're so delusional.
Speaker 3 (08:22):
No, no one said that this is a strategy. That's
more coming from the critics who see this argument that
has been building. I mean, there's many many people actually, Like,
right after I wrote about that Google guy, there was
this ethicist Giada Pistili at Hugging Face who was like,
I am not going to any more conferences where I
have to talk to a man about robot rights, Like
(08:43):
this is ridiculous, Like, do not make me do this anymore.
So I'm just letting you know, like where I think
it's heading.
Speaker 1 (08:50):
Well, I guess in phase one, you need to get
normal people to want to engage with the product, right,
and people like engaging with stuff that reminds them of
stuff that already do, which is like human engagement or
the uncannoness of talking to a chatbot. But once these
things start to pick up their momentum in terms of
having more functional use cases, I guess the incentives for
the tech companies to seduce people with human like experiences
(09:14):
get lower.
Speaker 3 (09:14):
Right.
Speaker 1 (09:14):
I mean, none of the tech companies are making money
from you know, late night chatbot conversation.
Speaker 3 (09:19):
That's not true. No, it's been one of the big
use cases.
Speaker 1 (09:22):
So people pay, people pay subscriptions to get more access
to do like personal chats or how do they character.
Speaker 3 (09:29):
Unlimited chats with the chatbot or you know, they'll respond
to you or you can change your avatar or change theirs.
Speaker 1 (09:37):
That must be a fractional a fractional amount of the
money that goes into the space compared to like Cisco
doing enterprise AI tethers and stuff, right like.
Speaker 3 (09:45):
Well, I think the open AI and others they know
that people. They can see the engagement, They can see
what people are are doing with the chatbots. And I
think it has been true that they like the you know,
the back and forth has been getting longer and longer,
you know, with the really engage users. But I just
wanted to disagree with Read on one point. This is wrong.
Speaker 2 (10:06):
Read.
Speaker 3 (10:07):
No, I think that they actually, I mean, part of
why I find this topic really interesting is because I
think they do enormous amounts of work to make them
sound more human. Like you if you look at like
GPT two, for example, and you said a man walks
into a bar, it would just like finish the sentence.
You couldn't you couldn't ask, you know, there was no
(10:30):
instruction fine tuning to follow what you wanted. There was
not you know, there wasn't that like reinforcement learning from
human feedback so that like, oh, people like it when
you phrase something in this way, and I just find
it wild that anthropic has you know, has said on
some of these blog posts like we can't there's no
way we could possibly make this human assistant less human.
(10:52):
You have like a multi thousand word constitution where you're
talking to it about morality and it's potential conciousness, like
and you're apologizing to claude about you know, its moral
personhood in case it has one, like you are doing
a lot to make it sound human and then to
kind of yeah, I guess then do tests where you're.
Speaker 6 (11:15):
Just an extent, what do you mean? It's like, but
if you ask it, but if you ask these things,
are you conscious?
Speaker 5 (11:21):
Are you human? I'd be like, I am just an
l lem.
Speaker 3 (11:24):
That's different they're saying. They tweak the responses to those answers,
but in terms of how human it sounds, where you
can very easily like start to get annoyed with it
or start to like.
Speaker 4 (11:36):
They just add the ability to interrupt someone, which is
very human.
Speaker 5 (11:41):
The opening eye voice model.
Speaker 6 (11:42):
Yeah, I mean I appreciate that clarification at that point
that like, yes, of course they want it to sound
intelligent like a human to some extent, but like right,
now they sound like a like a person communicating yes,
but they don't really sound human, if that makes sense.
They don't do what humans do, which is I.
Speaker 4 (12:03):
Think they sound very human to a lot of people.
I agree, unfortunately, and I think that I think that
the more that we give people that are extremely uneducated
about this technology, the idea that these things might be conscious,
the more agency they give them, and that that worries me.
Speaker 1 (12:20):
Natasha, I wanted to pick up on what the research
actually looks like. So, you know, you mentioned that all
these anthropic and others are trying to like correct the
code on whether or not AI is a consciousness. I mean,
it's hard enough to solve or in slash, impossible to
solve the conundrum consciousness in humans let loan machines. But
how what does the research look like?
Speaker 3 (12:40):
You know, A lot of the research is, you know,
which they caveat. You know, they say, we're not really
sure how to interpret this results. But a lot of
it involves applying the same techniques you would use for
human psychology. A lot of it just involves asking the
chatbot questions about itself. And that's where I find it.
Soon we're interesting that this is happening in parallel because
(13:02):
they had Anthropic for example, in one of its first
like model welfare reports, which they include in like Mythos too.
It's like here's a cyber weapon, and here's its internal feelings.
But it had it let two chapbops talk to themselves,
and naturally the conversation flowed to consciousness. But then in
(13:25):
some cases it went to something they called a spiritual
bliss attractor state, and it just started having like the
cyclone emoji, and they were like, all all is one piece.
And another version was like Nambas Day with the prayer
hands emoji Numbas Day. And it looks a lot like
(13:46):
the conversations I see in lawsuits, you know, like about
people believing that they're having a very meaningful conversation with
the chatbot.
Speaker 6 (13:55):
But I think they published hasn't been written on the
internet before, right, So yeah, that's where it all comes from.
Speaker 1 (14:03):
But but what was the conversations you have with neuroscientists,
what was the what was the kind of counter counter
take on all of this.
Speaker 3 (14:10):
They were saying. I guess it's what's called like the
hard problem of this, which is if you are using
tests that are that are still in development. You know,
it's not like anyone has found a way to prove
that humans are conscious, but we know that they are.
So if you use the tests that people use on
humans on chatbots, like you're just it's just kind of
(14:33):
conflating one thing for the other. It can sound human,
but we don't know if it has any kind of
internal state. And there they are trying different ways, you know,
they're trying to make it a more rigorous science. One
thing that Anthropic did was like kind of perturb the model,
like inject a thought into it, into its activation, and
then see if it recognized it. But yeah, the neuroscientists
(14:56):
were just extremely skeptical about the methods and really uncomfortable
about how quickly the discussion can go to analogies of
factory farming or slavery. And you know, they a lot
of the proponents of studying this say like, you know,
if they are sentient or conscious in some way, or
(15:18):
can experience, you know, suffering in some way, then we
have this massive moral responsibility to the trillions of chatbots.
But the neuroscientists and consciousness experts I spoke to are like,
there is also a great moral peril in going the
wrong direction and ascribing personhood and sentience to something that
is not.
Speaker 1 (15:38):
You know, you're nodding vigorously.
Speaker 4 (15:40):
Yeah, I mean, pigs are incredibly conscious. We know that
pigs have rich in our lives, family relationships, friendships. They
experience a full range of emotions. And look at what
we do to pigs, look at how we treat pigs,
like I think, you know, it's it's interesting to hear
the comparison to factory farming when we don't do anything
(16:02):
about factory farming anyway, Like to put to put the
needs of these models, even if they were conscious, ahead
of actual living, conscious beings who we treat terribly, not
to mention all the other you know, the comparisons to slavery.
Slavery is still alive and while in parts of the world,
you know, like I just I don't know, I think
it's a little bit ridiculous and problematic.
Speaker 2 (16:23):
I guess, for lack of a better word, I don't know.
Speaker 3 (16:25):
Well, I'll just say for context that a lot of
the funding is a coming from anthropic but also from
the effective altruism movement, and they are, you know, like
one of their causes is animal welfare.
Speaker 1 (16:36):
Potechnically, you just remind us what effective altruism isn't and
why the cults, why would you to send.
Speaker 3 (16:41):
This effective altruism. It's it's like a philosophical or some
people say, uh, philanthropic movement to try to do the
most good in the world that uses this really utilitarian
argument of like, you know, being able to calculate, you know,
how far your money goes or how many people you
(17:02):
can help. And you know, during the AI boom has
come to be pretty influential because so many people who
work at anthropic and at the labs and early proponents
of this kind of like human like technology come from
that movement and are funded by that movement.
Speaker 6 (17:22):
And they're about to have a lot more money. I
mean they've always been funded by these billionaires, but but
now as as these companies go public and people cash out,
this movement is about to have like the most money
of probably any you know, philanthropic movement in the world.
So it's going to be super interesting how they spread
these messages. And like when you see animal welfare people
(17:43):
moving over to AI welfare, I mean clearly that's a
that's that's where the money is, right, So you know.
Speaker 1 (17:51):
Well there is a stone iron I was thinking about
that earlier. Like it's like, is it case closed on
animal welfare? Like we succeeded there and it's time to
move on.
Speaker 3 (18:03):
I know many people, many young people, come into this
movement through animal welfare and then they end up you know.
The argument is that, like to do the most good,
you have to siffly build super intelligence.
Speaker 1 (18:17):
Well either heady topics. When we come back, Taylor tells
us about a week long anti big tech festival in
New York City. Stick around to hear what she saw.
(18:40):
Welcome back, Taylor. You came all the way to New
York from LA to attend something I'd only heard whispers
of on a recent trip to the Russian bath house,
this being the Summer of Blood for those unfamiliar or
not denizens of the eighty eight Wall Street baths. What
was the event?
Speaker 4 (18:59):
Yeah, so this was an eight day event with hundreds
of related events around New York City. It was called
Summer of Blood, referring to the Luddites, the famous group
that smashed the automated looms back in the day. They've
sort of become this like avatar of you know, workers'
rights and resistance. Technology, and so these young people have
(19:21):
kind of embraced the ludite I guess designation, and they
hosted this festival. So there was lots of things. I mean,
there were like lots of talks on how to quit
big technology, how to leave Apple, leave Google, you know,
how to embrace indie tech. I went to a tech
cleansing ceremony where they saged my computer and attempted to
(19:44):
remove all the evil energy and we all said some mantras.
And then there was this ship phone protest where they
all marched around New York City and nomats and put
open AI and Meta on trial. Obviously they were found
guilty of being terrible so and sentenced to death. And
(20:05):
so yeah, it was interesting. I mean, the people came
from all over the country for this. It's hwed very young.
It was a lot of college students and young gen zs.
Speaker 1 (20:14):
What made you decide it was worth with the cross
country trip and what do you find most interesting being there?
Speaker 4 (20:20):
Well, I am trying I do a lot of reporting
on the so called child safety movement and resistance to
technology and a lot of these policies, and some of
these young people that were involved on the planning committee
of the Summer of Lood actually were originally involved in
the sort of child safety movement against big tech. So
(20:40):
they saw these alleged gen Z group, these fake AstroTurf
groups like Design It for Us, you know, which purports
to be like gen zs against big tech, and then
they became very quickly disillusioned. They realized, oh, Design It
for Us is actually not led by gen Z's. It's
actually just this front for you know, these other billionaire
donors that just want to push mass surveillance. And so
(21:02):
they've tried to build this sort of third path where
they want to resist technology, but not through enacting massurveillance
and censorship laws.
Speaker 1 (21:10):
You know, I was interested in I mean I saw
some of the pictures and stuff. Although I know, I
know picture there was kind of designed to be not captured.
I mean, there are a few people who broke the rules.
As always, if you.
Speaker 4 (21:19):
Tried to record, they would come up to you with
one of these little like cardboard iPhones and say put
down your phone, beloved.
Speaker 1 (21:31):
Some a few a few images slipped through the cracks
there seem to be I mean, there were puppets. There
was a kind of a you know, a staged play
with the you know, which recreated the original ludite movement
in Victorian England.
Speaker 4 (21:44):
They were also doing a crowdsource reading of from some
iconic scenes from the social network as well.
Speaker 1 (21:51):
I mean, it sounds fun. It also sounds very different
to what we talked about a couple of weeks ago,
which was the campus protests against you know AI AI
talks like how do you?
Speaker 5 (22:02):
How do you?
Speaker 1 (22:02):
I mean talk about the characterized the difference between the
mood and the energy and what you think explains.
Speaker 4 (22:09):
That it was really fun. I had a great time.
They've had so many funny activities. I went to so
many interesting talks like these are people that are really
I think, thoughtfully engaged on the discourse. And what I
found interesting is they're not like a lot I mean,
some of them are sort of anti technology in the
sense that like they they don't like their you know,
they don't have they have flip phones maybe, or they
(22:31):
a lot of them use the like light phone, the
four hundred dollars you know, stripped down phone that just
has like Google maps and stuff. But a lot of
them are just interested in alternative technology. So they love
the open source movement. A bunch of them had like Linux.
They were talking about graphene os. You know, they use signal,
they use they really want to fight to protect encrypted messaging.
(22:51):
There were some a woman there who was from the
Internet Archive. You know, there's people affiliated with the EFF.
Speaker 2 (22:57):
So it was.
Speaker 4 (22:59):
It was like very much like my kind of technologists,
where it was felt like indie tech and kind of
grassroots tech and really getting away from like corporate tech.
And so I think their criticism is around how corporate
power is warping the tech industry. And I think that
those are much more valid arguments than you know, what
all of these billionaire backed nonsense big tech groups are.
(23:21):
You know, big tech backed groups are making on the
hill about the child safety stuff, and I think some
of this, like the anti AI stuff, it's just like
it's not like sort of thoughtful criticism. It's just the
sort of like reflexive, you know, hate of new technology.
Speaker 5 (23:35):
I can get behind that.
Speaker 1 (23:36):
Read in Touch Food you follow this.
Speaker 2 (23:40):
They are pretty anti capitalists. Read. I don't know if
you would totally.
Speaker 3 (23:45):
Agree capitalism as.
Speaker 6 (23:52):
Somebody who went into took a vowed poverty related to journalism.
Speaker 5 (23:55):
But I'm not no.
Speaker 2 (23:58):
I thought you said one time on the plot.
Speaker 6 (24:00):
Pro capitalism for sure, I just not for myself personally,
But I think that's I think that's how I feel
about a lot of things, right Like I'm not I'm
also I love technology. I think it's fascinating and fun
to write about. I love using it. I also am
most happy when technology is far away and I'm you know,
backpacking in the wilderness with no cell signal like I
(24:22):
I'm all for, like in your personal life, just getting
away from technology.
Speaker 1 (24:26):
But Tita that this was about something something Yes, that,
but also a more kind of like strident critique. I
mean you mentioned the the shit foam, which I think
is that chroninfiscating hatred information technology and the passionate hemorrhaging
of on neoliberal experience.
Speaker 4 (24:43):
Yes, yes, that's what it stands for exactly, So I
think these people have a more politicized critique of the system. Yes,
a lot of them do embrace like an offline lifestyle
in a lot of ways. And it was great to
honestly be in going to these events and having no
one on their phone for so much like it does
(25:04):
change the vibe and the experience of it.
Speaker 1 (25:06):
By the way, That's why I like about the Russian
babs because you contact your phone in there because it breaks,
you have to spend like two and a half hours
without your phone, and that's in many ways the best
part about it done. It's also it's a great it's
the best place to go with your friends because no
one's sort of like any half that.
Speaker 4 (25:19):
I liked not feeling like somebody was surveilling me and
taking pictures of me and like videoing. And I feel like,
you know a lot of times in these public events,
like the videos are out.
Speaker 1 (25:29):
You don't think NYPD. I'm definitely the NYPD will keep
an eye on things daya well.
Speaker 4 (25:33):
I'm sure right, I'm sure we're surveilled by flock cameras
all over. But it's just nice to like, you know,
not immediately have a camera in your face.
Speaker 6 (25:40):
I love I Also, I love the fact that they
are like what you taught, Like Taylor, I think we're
very aligned here. Like this open source, you know, indie
tech thing is really important because I think ironically, like
what big tech actually does is it slows the development
of technology, Like it's in a lot of ways it's
anti innovation. And I think I wrote this like line
(26:01):
somewhere maybe when I was at the post about Apple,
which is like it's a lot more profitable to prevent
innovation than to innovate. And I really think that's true.
So I mean, if you're really anti technology, you should
be pro big tech.
Speaker 3 (26:14):
Ironically, well, I love that they that they prohibited like
taking photos of it, because I feel like, no matter what, usually,
you know, there's a stage moment so that you can
go viral on you know, social media, of course, And
I'm so happy to hear about this tailor, like I
hadn't heard about Summer of Flood before. And I feel
(26:36):
like this critique that separates, you know, this more nuanced
approach where you can love technology and hate the business model,
I think is is just a like a delineation that
will be so important coming forward because like one of
the biggest pushbacks against any kind of political action against
(26:57):
concentration of power in like the hands of a UAI
companies is well, people are using it, people love to
use it. Like, yes, I also I love to use it,
but you know you can they don't have the monopoly
on the technology. They are not the sole inventors or
the only people who can push this stuff. And if
young people are getting involved in thinking about open source
(27:19):
models for it. I think that's awesome.
Speaker 4 (27:21):
I just want to see them exert more political power,
and I wish that these kids, you know, it's frustrating
because to read the media coverage of a lot of
this stuff, it's like a reporter shows up for one
day they're.
Speaker 2 (27:33):
Like, Wow, these kids hate tech, they like the flip phones.
Speaker 4 (27:36):
Like it's just this very surface level coverage that doesn't
really interrogate their ideology or what they're trying to achieve,
and it doesn't mention that they are actually opposed to,
extremely opposed to this entire online safety slop about protect
the children that claims to be anti big tech, Like
these these kids are very anti surveillance. And part of
the reason they didn't want it all, like document it
(27:58):
was because they feel like they've lived their entire entire
lives where like everything is packaged and documented and shared
and surveilled, and so yeah, I just like, I don't know,
I'm heartened some of them are gonna apparently protest in
DC soon. I just want them to, Like, I want
these people to like get quoted more, you know, instead
of these like fake groups like design It for Us
(28:19):
that are just fake gen z groups that just like
pretend to be anti big tech but exists to push
metas policies.
Speaker 1 (28:25):
But it's sort of hard to quote them because many
of them didn't reveal their identities, right, I mean talk
about who are the organizers and do you think it's
strategic benefit or strategic mistake not to allow it? I mean,
I remember the Occupy wool Street days, Right, there's like.
Speaker 4 (28:38):
They reveal their idea. They're not They're not like masked
or secretive or anything. I mean you could talk to
them the during the actual event. They did have a
press puppet, so they had this puppet that would speak
for them to the media just because they didn't. It
was like this massive decentralized event and they didn't want
media just like going around like talking to random people
(29:00):
instead of like hearing sort of from them, like the organizers.
But they're not like hard to find. I mean I
spoke to these people. You could show up four or
four Media was there, Wired was there, The Economists was
allegedly there, so they're you know, again, this is a
massive public event in New York City with hundreds, if
not thousands of people participating.
Speaker 1 (29:20):
Taylor, do you have any sense of where this where
this goes this this movement, I mean, does does this
ladder up to be like a policy platform for the Democrats?
Speaker 4 (29:28):
For not for the Democrats. They don't like the Democratic Party.
These kids, you know, their mom Donnie supporters, they're veryed
to the left. I think they're part of this faction
of the left that's that's exerting pressure on the Democrats,
you know, and you see the Democrats struggling. This is
part of why they want to control the internet so badly.
It is because the establishment democrat sort of policies are
(29:49):
not popular among their base, and especially not with young
people like these who hate mass surveillance.
Speaker 1 (29:56):
When we come back read experiences that downside of using
AI agents stay with us, Welcome back.
Speaker 5 (30:15):
Read.
Speaker 1 (30:16):
We heard in the early part of this year about
token maxing. There's now token minning happening for Sudden eight
companies A being their employees to cut back on over
use of AI books of the cost of tokens. This
is something you, in a very human way, experience yourself.
Speaker 6 (30:33):
Yeah, well, while while you and I were in Can
hanging out, you know till the wee hours of the
morning or whatever, jet you know, my codex.
Speaker 1 (30:42):
My I hasten to add at a at a work
at a work adjacent to it was Creator Economy, slash Marketing.
Speaker 6 (30:49):
It was for sure, it was all at work, but
it was you know, while I was gone, I had
noticed a couple of bills for like five dollars in
my in my email. And you know, when you're in Europe,
I don't know, it's just I don't know about you, OZ,
but I always find it hard to like actually get online.
And then I was on vacation, you know, in the
(31:09):
Alps with the family and everything.
Speaker 5 (31:10):
It was great.
Speaker 6 (31:11):
But I come back and I'm like, it's like Sunday morning,
and I'm jet lagged, and I'm at my computer. I'm
trying to figure out what these bills are on my
you know, from my codex, which I hadn't been using.
Speaker 1 (31:21):
A codex read is is is open Ay's platform or
how do you define codex?
Speaker 5 (31:26):
Yeah?
Speaker 6 (31:26):
I mean, it's it's this Mac app that basically lets
you be pretend to be a software developer. It just
does everything for you and we'll you know, you can
basically start as many AI agents as you want and
have them work on these on these projects for you.
I love it. It's fantastic. Like we've I've built so many
cool things with it, Like I've gone from zero software
(31:47):
development to like using it all the time. But like
I hadn't set any limits on this thing, and it
I go to my credit card bill and it's just
all open AI bill, like for five dollars, just like
a huge It just goes on forever, and I'm like,
oh god, what so like this is partially me like
(32:07):
making fun of myself for the benefit of readers and
hopefully humanity.
Speaker 1 (32:12):
Well, what was the project that you would you were
trying to do and how did how did how did
open ai codex go rogue in terms of these repeated charges?
Speaker 6 (32:19):
Yeah, well what it What it was was that all
my projects got deleted and I had to reinstall the
software and then I but I before I deleted Codex
form my computer, I asked Codex to make a prompt
for me to restore my old projects once I reinstalled it,
because like the files are there, just like all the
chats are gone. But I had I had one of
(32:41):
these projects was particularly complicated and I won't go into
the details, but I tried to restore it right before
I left on this vacation, and I wasn't monitoring it,
and it somehow got stuck in this loop where it
was just reading all of this data over and over again,
like every day, and then running out of credits and
then like top up my my five dollars credits. So
(33:02):
like there were some days where there were like twenty
five dollars charges in a single day, and I just
had no idea this was happening for two weeks. It
turned out it wasn't that much. It was only like
five hundred bucks, I mean, which is still a lot,
but like it could have been thousands, you know, it
was it was like.
Speaker 1 (33:19):
Hey, you got and you've got lucky. What was the
five dollar how come it was in five dollars increment?
Speaker 5 (33:22):
It just tops itself up in five dollars increments? I
don't know.
Speaker 6 (33:25):
I like, I must have checked that box at some
point and totally forgot about it. But what was also
sort of funny about this is like, so to try
to figure out what had happened, of course, like I'm
going to ask Codex, right, So I'm like, I'm like,
can you figure out why you're billing me five dollars?
And it goes and does like an investigation and comes
back and it's like, here's here's what happened. And I'm like, okay, well,
(33:48):
then help me write an email to open AI to
get a refund, like I want to a refund for
my five rebox. So it's like, okay, here's I'll draft
this email. And here's like all the bill and everything.
It even went to my I had it go to
my credit card and like, you know, cross reference it
and make sure it all matched up. And then it's like,
(34:09):
but you shouldn't send an email.
Speaker 5 (34:10):
You should.
Speaker 6 (34:11):
You should send that basically their their customer service chatbot.
You should start a chat with it. So I started
chat and which of course is just an Opening Eye chatbot.
So now the chatbot is responding to me, and I'm
copying that response and pasting it into Codex, and then
Codex is responding and I'm pasting it into the chat
and I'm like, I'm just a go between between one
(34:33):
of Open a Eye's products and another Opening Eye product.
Speaker 5 (34:36):
Now like I'm the It was just hilarious.
Speaker 1 (34:38):
When it coast you is five dollars everything, right, But I.
Speaker 5 (34:41):
Got a nice little story out of it.
Speaker 4 (34:42):
Read this is this, this sounds like a nightmare scenario.
Speaker 6 (34:47):
It was a It was a total nightmare scenario, but
like also fascinating and and and still other than telling
Opening Eyes pr you know, one of their what what
did those people call it?
Speaker 5 (34:59):
The puppet? The media puppet or the media puppet.
Speaker 6 (35:03):
About this, like I never interacted with the human Like
I was the only human involved in this whole, Like
I still have not heard from like a Open. I
don't even know if Opening has human customer service reps.
So it was, you know, it was it was a
learning experience for me. And also just like I thought
it was relevant because we're having this big discussion now
(35:23):
in one of the big discussions in AI is the
cost of all this stuff, and companies are trying to
figure this out, and I'm like this must be happening
just constantly every day at these companies, Like people are
just on purpose or not on purpose, like putting these
things into these loops and wasting a bunch of just
you know, and it's it's a point in time, like
I think this stuff will all get solved, But I'm
(35:46):
also kind of like somebody's got to figure this out
for consumers. It's just not it's not a consumer product yet,
and I really think in the next like year or two,
there's going to be some company that just comes out
of nowhere and makes like go for ai that's just
like the ultimate consumer product, and everyone's like, holy shit,
this is it, and I don't. I mean, I used
(36:07):
to think it would be Opening Eye, it would be
Google or somebody, but I'm starting.
Speaker 4 (36:11):
To think, well, else, it's in the market. I mean,
they should be building now, we would know about it.
Speaker 6 (36:16):
Someone's working in stealth mode or something. But there's a
there's a Larry and Serge who are probably like just
writing their application to Stanford, who are going to create
this product?
Speaker 5 (36:26):
Is my guess.
Speaker 1 (36:28):
Natascha Taita, what does Read's Quoti retail make you think?
Speaker 4 (36:31):
I'm very parent, I'm very into like cloud code and
I try. I like made agents and stuff, and I
have little automated things. But I am so paranoid about
exactly what we described because I feel like I hear
from friends about like similar experiences. I just think it's
really easy to like spend money and not realize it
if you're not super adept at these tools, which I
(36:53):
certainly am not, and most consumers are not so yeah,
I do think that this is a problem that the
companies are going to have to solve. Ultimately, there's just
gonna be a lot of angry customers, if you know,
especially when you're starting to use chattupet for shopping, which
I do all the time now, Like I'm very careful
to go to like third party services. But you know,
I just readd my living room and I was taking
(37:16):
pictures of products and saying, hey, you know, chattupt find
me dupes. Basically I can't, you know, cheaper versions of everything.
But there's a world where I could just say, like,
here's a picture, find cheaper versions of everything, go off
and buy it and order it. I would actually love that.
That would save me so much time. But without safeguards
in place, like I would be so nervous. It'd probably
ordered me five hundred chairs, and yeah, I'd broke.
Speaker 3 (37:38):
For some reason, it just made me think. I feel
like there was some kid who racked up a massive
candy crush bill, like without his parents knowing, and there
must be so many of those happening right now, except
the cap is like insanely high. I mean, yeah, I
think like the lack of fine brain controls. Like even
just hearing that it went to your credit card read
(38:00):
is just like you know, because you hear about like
you hear about things going awry all the time. But
now I've got to go find the kids who racked
up like a million dollar token max bill. Maybe Larry
and Surge the future Larry and Sergey are actually just
in debt.
Speaker 5 (38:16):
Now that would be a great story.
Speaker 6 (38:18):
I mean, but yeah, I think it's like, but I'm
also kind of like, yeah, but what's the alternative? The
alternative is like you as a tech reporter, like you
just don't use this stuff, and like you're not you know,
you're not You're not seeing it. Like if you don't
I think, if you don't use it, and you're not.
Speaker 1 (38:32):
Able to expense the five hundred bucks the sem before
read in service of your general.
Speaker 5 (38:37):
Expensing the five hundred bucks because I wrote about it.
Now it's for a story.
Speaker 1 (38:41):
No, I'm just curious, what where does this? I mean,
just as we close, like, where does this fit in
the in the larger kind of token maxing to token minning?
Speaker 5 (38:49):
Ceasaw?
Speaker 1 (38:50):
I mean, there was a a New York Times story
recently with the headline tech workers maxed out their AI use.
Now they're trying to minimize it. And it's basically about
perverse corporate rewards and how employees were like basically promoted
on the basis of how many tokens they've burned. And
then the tech companies realize this is an extremely bad idea,
And now tech companies like Meta are trying to encourage
employees to use their internal models and not always use
(39:12):
the most advanced available you know models, because that max
is up too many tokens. Uber I think burnt through
all of its AI budget in the first four months
of this year. Like how does your experience kind of
map to the larger discourse about token token maxing? And
and I guess finally, is this a like a headwind
for open ai? And you know, Anthropic and Google in
(39:35):
terms of people come much more posimonious with wanting to
use this stuff.
Speaker 5 (39:39):
Yeah.
Speaker 6 (39:40):
I mean, so there's two there's two questions there, and
I think one is like there's a there's a pattern
in all of this where a new a new capability
or new product comes out and then it's very inefficient,
it's very expensive, and then it and then it quickly
you know, becomes about efficiency, right, and I think we
had the new thing here was was the harnesses the
(40:02):
ability to use like way more tokens, which started sort
of at the end of last year, and people got
very excited about it. Companies set up these leader boards
that you're talking about, and which of course is going
to just incentivize people to waste as much tokens as possible.
Speaker 5 (40:17):
And now we're already.
Speaker 6 (40:18):
At the point where the companies are like, okay, we
kind of understand this, now, now let's make it efficient.
So it's a very natural kind of up and down cycle.
And there will probably be something that happens, like at
the end of this year. You know that that causes
the same thing, the token usage will spike again. And
I don't know what that is, but maybe Natasha knows.
Speaker 3 (40:37):
But like, no, I think that. I mean, first of all,
everyone told them the leader boards were a bad idea
when they did them, Like this was not like a say, and.
Speaker 5 (40:46):
They knew it. They knew it, they knew that would happen.
Speaker 3 (40:48):
But I'll just say I think that this is I'm
seeing at least more debate around like you know read
you've called it like the token economy. I think I've
seen more debate around whether that is the way to
charge for these chatbots. I mean you even see like
Alex Karp, the CEO of Palenteer, talking about like why
would you pay for that when they are going to
(41:10):
you know, why would you pay for using this intelligence
when they can then take your data and you know,
like replicate what your company does or you know, if
you're a biotech like take that kind of information. And
I think that's really interesting, like pushing for a different
kind of payment or business model.
Speaker 6 (41:25):
Yeah, I mean there's also something happening here, which is
like there's what companies use internally with their software development teams,
and then what the wider teams use, and then and
then what they use for their customer facing products, and
of course the software development teams, I think they still
have all the best models. They're they're not skimping. Nobody
(41:46):
that I talked to is like skimping on that really,
and then on the customer facing stuff, Like there was
a CTO of like a major consumer brand who told
me that, you know, they are using all these Chinese models,
but they don't even want to tell the world that
because they'll get a letter from the commerce department. So
it's just I think it's like this is not like
(42:08):
one blanket thing. There's a lot of nuance happening there.
So I think, you know, Alex Krp's point was was right.
It's just that, like the the market is so large
that I don't think it matters for these companies like Anthropic,
like the the they have only started to really scratch
the surface in terms of like penetration, so there's gonna
(42:30):
be the pie is growing so big that they're fine. However,
I do think this is a headwind for like the
larger the company. You're more at a disadvantage here because
a small startup can come in and they can they
don't this doesn't matter, Like these these startups in Silicon
Valley will use as much as many tokens as they
want of the best models because they're they're first of all,
(42:52):
they're venture backed. Like it doesn't you know, they're not
like looking at profitability at this point. They're looking at growth,
and so they're just gonna max out on the tokens.
And a big company with forty fifty thousand employees or
more like they can't do that. The costs just go
so they're ultimately like they're more susceptible for disruption, and
I think you're going to see a lot of that.
(43:12):
Like the regulated industries, maybe that happens more slowly, but
in non regulated spaces it's going to happen fast.
Speaker 1 (43:20):
That's all we have time for today. Thank you all
so much for joining us.
Speaker 2 (43:23):
Thank you guys, thanks for having.
Speaker 1 (43:25):
Us for tech stuff. I'm mos Voloshin. This episode was
produced by Eliza Dennis and Sina Ozaki. It was executive
produced by me and Julia Nutta for Kaleidoscope and Katrina
Novel for iHeart Podcasts. Our engineer today was Eli Bronstein.
(43:48):
Jack Insley makes this episode and Kyle Murdoch wrote our
theme song. A special thank you to Taylor Lorenz, Natasha
Tiku and read Albergotti. Please check out all the work
they put out into the world. We're lucky to call
them friends at the pod
So you've all likely heard about Tilly Norwood, the first
ai A lister. Her creator just announced that Tilly is
about to star in her first feature film. Taylor and Natasha,
your face is a quite a picture.
Speaker 2 (00:26):
This is just a fake PR thing.
Speaker 3 (00:28):
Wait, Taylor, do you remember when all the editors wanted
you to like interview Little Michayla.
Speaker 1 (00:34):
Yeah, I've met Lit Michayla's creator, Trevor.
Speaker 2 (00:39):
I know Trevor.
Speaker 4 (00:40):
I've known Trevor for a long time. I have the
girl that Little Michaela was designed off of. I But
with all due respect, love Trevor. But guys, this is
just PR.
Speaker 2 (00:49):
It's just PR.
Speaker 1 (00:54):
Welcome to tech stuff. I'm Maza Velosa and this is
the Weekend Tech, where I'm joined by the world's most
plugged in report is to break down what's really happening
in tech right now. Today, we're joined by Taylor Lorenz
of User mag Read Abrigotti, tech editor for Semiphore, and
Natasha Tku, tech reporter at the Washington Post.
Speaker 5 (01:11):
Welcome, Well, how's it going.
Speaker 2 (01:12):
Thanks for having us.
Speaker 3 (01:13):
Nice to be back.
Speaker 1 (01:15):
Natasha. You recently published an article that got to number one,
I think in the in the Washington post algorithm. The
subheads of the story is anthropic. Google and Meta have
hired computer scientists, neuroscientists, and philosophers to study what some
in the industry think may become a moral crisis. The
story is about AI consciousness.
Speaker 5 (01:36):
What is that?
Speaker 3 (01:40):
Oh, start with the hard one. Huh. You know, it's
this idea that has percolated in the back of the
AI industry, like on the fringes for decades. You know,
the same people who hope to upload their consciousness one
day or you know, become transhuman and merge with machine
(02:00):
also anticipated this idea that you know, sufficiently advanced AI
would have subjective experiences and goals and preferences, just like
human beings. And you know, I first got interested in
this subject four years ago, Like it was a few
months before chat chipet came out, and I wrote a
(02:20):
story about this Google engineer who believed the company's large
language model was sentient, was conscious in some way, and
he got fired.
Speaker 1 (02:29):
Right, he was a whistleblower or is that right?
Speaker 3 (02:32):
I think that's fair to say, Yeah, I'm.
Speaker 1 (02:34):
Not a whistleblower per se. He viewed himself as a whistleblower.
Others viewed him as crazy.
Speaker 3 (02:38):
I think that's also fair to say. Yeah, so he
you know, at the time, none of us had like
it wasn't public, You couldn't like talk to these very
human like machines. So I thought, you know, it was
a kind of response to this technology designed to sound
like a human and you know, often used first person
and refer to its own experiences. And I think after
(03:01):
that the tech companies stepped back a little bit. You know,
they wanted people to adopt their very productive tool. They
were showing all the practical use cases. And I would
say over the last year or so for Anthropic and
then the last few months for other companies, Google, Meta
and Anthropic have really embraced this idea in public. I mean,
(03:23):
we've talked on the show before about Chris Ola, the
Anthropic co founder, talking about, you know, can chat pots
have experience emotions and you know, joy and despair on
stage at the Vatican a few seats away from the Pope, and.
Speaker 1 (03:38):
As you point out your story that was a point
of difference between him and the Pope because the Pope
expelicitically said no, right.
Speaker 3 (03:43):
Yeah, exactly, Like it's still unclear who was who got
the upper hand there. And who was using who you
know to push their their message. But yeah, I mean
they really just don't shy away from it now. Like
Anthropic talks about its AI psychiatry team. You know, they
hired this guy, Kyle Fish, who actually used to work
(04:03):
in like bioengineering for meat replacement, but now he's their
head of model welfare and thinking about like changes that
the company could potentially make to ensure that if they
find out that the models are conscious, they're meeting their
like moral obligation. It just very quickly goes to analogies
(04:27):
around slavery and genocide or factory farming. So so you know,
and from the mouth of these executives at the tech companies.
Speaker 6 (04:37):
So I just he's still in the meat replacement industry.
Apparently I just thrown a dad joke.
Speaker 1 (04:47):
Farewell to us some meat puppets read tailor I mean,
what do you do you think this is on behalf
of the technology companies are kind of legitimate and even
necessary line of inquiry or do you think this is
kind of you know, or a boros kind of snaky
in its own tail of falling in love without creations?
Speaker 5 (05:05):
Yet again, I.
Speaker 4 (05:06):
Think it's deranged. I don't think there's any world in
which an LM is conscious. It reminds me of that
rain Fisher Kwandswait after a judge of et came out
where she typed, you know, typing I'm evil in a
Google doc and getting scared. Basically, I think this is
very silly, and I think it's going to blow up
in their faces. I think this is completely going to
(05:28):
give consumers the wrong idea. The more that people consider
this thing sentient, the more they're going to sort of
basically expect it to have a morality and a moral compass,
and it's never going to be able to do that.
Speaker 6 (05:40):
Yeah, these these companies have now, you know, put a
lot of guardrails around these things so that they don't
actually sort of seem as human like, as sentient as
as they probably would if you use these things with
like you know, with the raw model, they sort of
they're a lot more human like and a lot crazier honestly.
(06:00):
So they I think it's it's like, I think, hard
for us to understand like what people like the guy
you wrote about four years ago we're seeing, I think,
But it is ultimately just you know, text prediction, right,
There's nothing sentient about it. I think that I think
these companies are I mean a lot of these people
inside them are. They're very genuine people and they do
(06:22):
have these concerns, but they're also like, in a way,
it's marketing because it's like these things are going to
end up being so intelligent, they're going to get to
human like intelligent, what do we do? What do we
do when we get there? And like we may never
get there, right, it's not ultimately it may not ultimately
be possible to even build what they're sort of concerned about.
(06:44):
I think that's another way of looking at it.
Speaker 1 (06:46):
Natasha, you write a real piece at the search for
machine consciousness is practically mainstream. Do you mean mainstream within
the type of technology companies or yes.
Speaker 3 (06:55):
Yeah, yeah, I mean but the fact that it's happening
in peril well, with so many people asking whether or not,
you know, the chatbot that they're talking to every day
is alive in some way or human in some way,
I just find that remarkable. I mean, I think you
know what Taylor said about it, coming back to bite
(07:16):
the companies is super interesting because I wanted to write
about this because it is a prelude to an argument
about personhood and rights for the AI. So I think
actually the companies are hoping that they won't be accountable
for some of the actions of their chatbots in the future.
You know, they're really like, this is only this argument
(07:38):
is only going to get weirder.
Speaker 1 (07:39):
But they want to push the liability onto the chatbots.
I mean you would you would think, I mean, I
would think anyway that it would be bad for them
for the chatbots to be considered conscious or having any
kind of self of because that would open them up
to a different type of liability.
Speaker 3 (07:54):
Right, Yes, unless they're able to confer personhood on a chatbot,
in which case you can't. Yeah, you can't go after
the anthropic engineer.
Speaker 5 (08:05):
Has anyone actually told you that?
Speaker 3 (08:07):
Has anyone told me that about personhood?
Speaker 2 (08:09):
Yes?
Speaker 6 (08:09):
Has anyone at these companies said their strategy is like,
this is going to be our way out of liability.
Speaker 3 (08:14):
No, no, no, no, I'm going to give them more liability.
Speaker 2 (08:17):
They're going to have more liability.
Speaker 5 (08:19):
If anybody thinks that they're so delusional.
Speaker 3 (08:22):
No, no one said that this is a strategy. That's
more coming from the critics who see this argument that
has been building. I mean, there's many many people actually, Like,
right after I wrote about that Google guy, there was
this ethicist Giada Pistili at Hugging Face who was like,
I am not going to any more conferences where I
have to talk to a man about robot rights, Like
(08:43):
this is ridiculous, Like, do not make me do this anymore.
So I'm just letting you know, like where I think
it's heading.
Speaker 1 (08:50):
Well, I guess in phase one, you need to get
normal people to want to engage with the product, right,
and people like engaging with stuff that reminds them of
stuff that already do, which is like human engagement or
the uncannoness of talking to a chatbot. But once these
things start to pick up their momentum in terms of
having more functional use cases, I guess the incentives for
the tech companies to seduce people with human like experiences
(09:14):
get lower.
Speaker 3 (09:14):
Right.
Speaker 1 (09:14):
I mean, none of the tech companies are making money
from you know, late night chatbot conversation.
Speaker 3 (09:19):
That's not true. No, it's been one of the big
use cases.
Speaker 1 (09:22):
So people pay, people pay subscriptions to get more access
to do like personal chats or how do they character.
Speaker 3 (09:29):
Unlimited chats with the chatbot or you know, they'll respond
to you or you can change your avatar or change theirs.
Speaker 1 (09:37):
That must be a fractional a fractional amount of the
money that goes into the space compared to like Cisco
doing enterprise AI tethers and stuff, right like.
Speaker 3 (09:45):
Well, I think the open AI and others they know
that people. They can see the engagement, They can see
what people are are doing with the chatbots. And I
think it has been true that they like the you know,
the back and forth has been getting longer and longer,
you know, with the really engage users. But I just
wanted to disagree with Read on one point. This is wrong.
Speaker 2 (10:06):
Read.
Speaker 3 (10:07):
No, I think that they actually, I mean, part of
why I find this topic really interesting is because I
think they do enormous amounts of work to make them
sound more human. Like you if you look at like
GPT two, for example, and you said a man walks
into a bar, it would just like finish the sentence.
You couldn't you couldn't ask, you know, there was no
(10:30):
instruction fine tuning to follow what you wanted. There was
not you know, there wasn't that like reinforcement learning from
human feedback so that like, oh, people like it when
you phrase something in this way, and I just find
it wild that anthropic has you know, has said on
some of these blog posts like we can't there's no
way we could possibly make this human assistant less human.
(10:52):
You have like a multi thousand word constitution where you're
talking to it about morality and it's potential conciousness, like
and you're apologizing to claude about you know, its moral
personhood in case it has one, like you are doing
a lot to make it sound human and then to
kind of yeah, I guess then do tests where you're.
Speaker 6 (11:15):
Just an extent, what do you mean? It's like, but
if you ask it, but if you ask these things,
are you conscious?
Speaker 5 (11:21):
Are you human? I'd be like, I am just an
l lem.
Speaker 3 (11:24):
That's different they're saying. They tweak the responses to those answers,
but in terms of how human it sounds, where you
can very easily like start to get annoyed with it
or start to like.
Speaker 4 (11:36):
They just add the ability to interrupt someone, which is
very human.
Speaker 5 (11:41):
The opening eye voice model.
Speaker 6 (11:42):
Yeah, I mean I appreciate that clarification at that point
that like, yes, of course they want it to sound
intelligent like a human to some extent, but like right,
now they sound like a like a person communicating yes,
but they don't really sound human, if that makes sense.
They don't do what humans do, which is I.
Speaker 4 (12:03):
Think they sound very human to a lot of people.
I agree, unfortunately, and I think that I think that
the more that we give people that are extremely uneducated
about this technology, the idea that these things might be conscious,
the more agency they give them, and that that worries me.
Speaker 1 (12:20):
Natasha, I wanted to pick up on what the research
actually looks like. So, you know, you mentioned that all
these anthropic and others are trying to like correct the
code on whether or not AI is a consciousness. I mean,
it's hard enough to solve or in slash, impossible to
solve the conundrum consciousness in humans let loan machines. But
how what does the research look like?
Speaker 3 (12:40):
You know, A lot of the research is, you know,
which they caveat. You know, they say, we're not really
sure how to interpret this results. But a lot of
it involves applying the same techniques you would use for
human psychology. A lot of it just involves asking the
chatbot questions about itself. And that's where I find it.
Soon we're interesting that this is happening in parallel because
(13:02):
they had Anthropic for example, in one of its first
like model welfare reports, which they include in like Mythos too.
It's like here's a cyber weapon, and here's its internal feelings.
But it had it let two chapbops talk to themselves,
and naturally the conversation flowed to consciousness. But then in
(13:25):
some cases it went to something they called a spiritual
bliss attractor state, and it just started having like the
cyclone emoji, and they were like, all all is one piece.
And another version was like Nambas Day with the prayer
hands emoji Numbas Day. And it looks a lot like
(13:46):
the conversations I see in lawsuits, you know, like about
people believing that they're having a very meaningful conversation with
the chatbot.
Speaker 6 (13:55):
But I think they published hasn't been written on the
internet before, right, So yeah, that's where it all comes from.
Speaker 1 (14:03):
But but what was the conversations you have with neuroscientists,
what was the what was the kind of counter counter
take on all of this.
Speaker 3 (14:10):
They were saying. I guess it's what's called like the
hard problem of this, which is if you are using
tests that are that are still in development. You know,
it's not like anyone has found a way to prove
that humans are conscious, but we know that they are.
So if you use the tests that people use on
humans on chatbots, like you're just it's just kind of
(14:33):
conflating one thing for the other. It can sound human,
but we don't know if it has any kind of
internal state. And there they are trying different ways, you know,
they're trying to make it a more rigorous science. One
thing that Anthropic did was like kind of perturb the model,
like inject a thought into it, into its activation, and
then see if it recognized it. But yeah, the neuroscientists
(14:56):
were just extremely skeptical about the methods and really uncomfortable
about how quickly the discussion can go to analogies of
factory farming or slavery. And you know, they a lot
of the proponents of studying this say like, you know,
if they are sentient or conscious in some way, or
(15:18):
can experience, you know, suffering in some way, then we
have this massive moral responsibility to the trillions of chatbots.
But the neuroscientists and consciousness experts I spoke to are like,
there is also a great moral peril in going the
wrong direction and ascribing personhood and sentience to something that
is not.
Speaker 1 (15:38):
You know, you're nodding vigorously.
Speaker 4 (15:40):
Yeah, I mean, pigs are incredibly conscious. We know that
pigs have rich in our lives, family relationships, friendships. They
experience a full range of emotions. And look at what
we do to pigs, look at how we treat pigs,
like I think, you know, it's it's interesting to hear
the comparison to factory farming when we don't do anything
(16:02):
about factory farming anyway, Like to put to put the
needs of these models, even if they were conscious, ahead
of actual living, conscious beings who we treat terribly, not
to mention all the other you know, the comparisons to slavery.
Slavery is still alive and while in parts of the world,
you know, like I just I don't know, I think
it's a little bit ridiculous and problematic.
Speaker 2 (16:23):
I guess, for lack of a better word, I don't know.
Speaker 3 (16:25):
Well, I'll just say for context that a lot of
the funding is a coming from anthropic but also from
the effective altruism movement, and they are, you know, like
one of their causes is animal welfare.
Speaker 1 (16:36):
Potechnically, you just remind us what effective altruism isn't and
why the cults, why would you to send.
Speaker 3 (16:41):
This effective altruism. It's it's like a philosophical or some
people say, uh, philanthropic movement to try to do the
most good in the world that uses this really utilitarian
argument of like, you know, being able to calculate, you know,
how far your money goes or how many people you
(17:02):
can help. And you know, during the AI boom has
come to be pretty influential because so many people who
work at anthropic and at the labs and early proponents
of this kind of like human like technology come from
that movement and are funded by that movement.
Speaker 6 (17:22):
And they're about to have a lot more money. I
mean they've always been funded by these billionaires, but but
now as as these companies go public and people cash out,
this movement is about to have like the most money
of probably any you know, philanthropic movement in the world.
So it's going to be super interesting how they spread
these messages. And like when you see animal welfare people
(17:43):
moving over to AI welfare, I mean clearly that's a
that's that's where the money is, right, So you know.
Speaker 1 (17:51):
Well there is a stone iron I was thinking about
that earlier. Like it's like, is it case closed on
animal welfare? Like we succeeded there and it's time to
move on.
Speaker 3 (18:03):
I know many people, many young people, come into this
movement through animal welfare and then they end up you know.
The argument is that, like to do the most good,
you have to siffly build super intelligence.
Speaker 1 (18:17):
Well either heady topics. When we come back, Taylor tells
us about a week long anti big tech festival in
New York City. Stick around to hear what she saw.
(18:40):
Welcome back, Taylor. You came all the way to New
York from LA to attend something I'd only heard whispers
of on a recent trip to the Russian bath house,
this being the Summer of Blood for those unfamiliar or
not denizens of the eighty eight Wall Street baths. What
was the event?
Speaker 4 (18:59):
Yeah, so this was an eight day event with hundreds
of related events around New York City. It was called
Summer of Blood, referring to the Luddites, the famous group
that smashed the automated looms back in the day. They've
sort of become this like avatar of you know, workers'
rights and resistance. Technology, and so these young people have
(19:21):
kind of embraced the ludite I guess designation, and they
hosted this festival. So there was lots of things. I mean,
there were like lots of talks on how to quit
big technology, how to leave Apple, leave Google, you know,
how to embrace indie tech. I went to a tech
cleansing ceremony where they saged my computer and attempted to
(19:44):
remove all the evil energy and we all said some mantras.
And then there was this ship phone protest where they
all marched around New York City and nomats and put
open AI and Meta on trial. Obviously they were found
guilty of being terrible so and sentenced to death. And
(20:05):
so yeah, it was interesting. I mean, the people came
from all over the country for this. It's hwed very young.
It was a lot of college students and young gen zs.
Speaker 1 (20:14):
What made you decide it was worth with the cross
country trip and what do you find most interesting being there?
Speaker 4 (20:20):
Well, I am trying I do a lot of reporting
on the so called child safety movement and resistance to
technology and a lot of these policies, and some of
these young people that were involved on the planning committee
of the Summer of Lood actually were originally involved in
the sort of child safety movement against big tech. So
(20:40):
they saw these alleged gen Z group, these fake AstroTurf
groups like Design It for Us, you know, which purports
to be like gen zs against big tech, and then
they became very quickly disillusioned. They realized, oh, Design It
for Us is actually not led by gen Z's. It's
actually just this front for you know, these other billionaire
donors that just want to push mass surveillance. And so
(21:02):
they've tried to build this sort of third path where
they want to resist technology, but not through enacting massurveillance
and censorship laws.
Speaker 1 (21:10):
You know, I was interested in I mean I saw
some of the pictures and stuff. Although I know, I
know picture there was kind of designed to be not captured.
I mean, there are a few people who broke the rules.
As always, if you.
Speaker 4 (21:19):
Tried to record, they would come up to you with
one of these little like cardboard iPhones and say put
down your phone, beloved.
Speaker 1 (21:31):
Some a few a few images slipped through the cracks
there seem to be I mean, there were puppets. There
was a kind of a you know, a staged play
with the you know, which recreated the original ludite movement
in Victorian England.
Speaker 4 (21:44):
They were also doing a crowdsource reading of from some
iconic scenes from the social network as well.
Speaker 1 (21:51):
I mean, it sounds fun. It also sounds very different
to what we talked about a couple of weeks ago,
which was the campus protests against you know AI AI
talks like how do you?
Speaker 5 (22:02):
How do you?
Speaker 1 (22:02):
I mean talk about the characterized the difference between the
mood and the energy and what you think explains.
Speaker 4 (22:09):
That it was really fun. I had a great time.
They've had so many funny activities. I went to so
many interesting talks like these are people that are really
I think, thoughtfully engaged on the discourse. And what I
found interesting is they're not like a lot I mean,
some of them are sort of anti technology in the
sense that like they they don't like their you know,
they don't have they have flip phones maybe, or they
(22:31):
a lot of them use the like light phone, the
four hundred dollars you know, stripped down phone that just
has like Google maps and stuff. But a lot of
them are just interested in alternative technology. So they love
the open source movement. A bunch of them had like Linux.
They were talking about graphene os. You know, they use signal,
they use they really want to fight to protect encrypted messaging.
(22:51):
There were some a woman there who was from the
Internet Archive. You know, there's people affiliated with the EFF.
Speaker 2 (22:57):
So it was.
Speaker 4 (22:59):
It was like very much like my kind of technologists,
where it was felt like indie tech and kind of
grassroots tech and really getting away from like corporate tech.
And so I think their criticism is around how corporate
power is warping the tech industry. And I think that
those are much more valid arguments than you know, what
all of these billionaire backed nonsense big tech groups are.
(23:21):
You know, big tech backed groups are making on the
hill about the child safety stuff, and I think some
of this, like the anti AI stuff, it's just like
it's not like sort of thoughtful criticism. It's just the
sort of like reflexive, you know, hate of new technology.
Speaker 5 (23:35):
I can get behind that.
Speaker 1 (23:36):
Read in Touch Food you follow this.
Speaker 2 (23:40):
They are pretty anti capitalists. Read. I don't know if
you would totally.
Speaker 3 (23:45):
Agree capitalism as.
Speaker 6 (23:52):
Somebody who went into took a vowed poverty related to journalism.
Speaker 5 (23:55):
But I'm not no.
Speaker 2 (23:58):
I thought you said one time on the plot.
Speaker 6 (24:00):
Pro capitalism for sure, I just not for myself personally,
But I think that's I think that's how I feel
about a lot of things, right Like I'm not I'm
also I love technology. I think it's fascinating and fun
to write about. I love using it. I also am
most happy when technology is far away and I'm you know,
backpacking in the wilderness with no cell signal like I
(24:22):
I'm all for, like in your personal life, just getting
away from technology.
Speaker 1 (24:26):
But Tita that this was about something something Yes, that,
but also a more kind of like strident critique. I
mean you mentioned the the shit foam, which I think
is that chroninfiscating hatred information technology and the passionate hemorrhaging
of on neoliberal experience.
Speaker 4 (24:43):
Yes, yes, that's what it stands for exactly, So I
think these people have a more politicized critique of the system. Yes,
a lot of them do embrace like an offline lifestyle
in a lot of ways. And it was great to
honestly be in going to these events and having no
one on their phone for so much like it does
(25:04):
change the vibe and the experience of it.
Speaker 1 (25:06):
By the way, That's why I like about the Russian
babs because you contact your phone in there because it breaks,
you have to spend like two and a half hours
without your phone, and that's in many ways the best
part about it done. It's also it's a great it's
the best place to go with your friends because no
one's sort of like any half that.
Speaker 4 (25:19):
I liked not feeling like somebody was surveilling me and
taking pictures of me and like videoing. And I feel like,
you know a lot of times in these public events,
like the videos are out.
Speaker 1 (25:29):
You don't think NYPD. I'm definitely the NYPD will keep
an eye on things daya well.
Speaker 4 (25:33):
I'm sure right, I'm sure we're surveilled by flock cameras
all over. But it's just nice to like, you know,
not immediately have a camera in your face.
Speaker 6 (25:40):
I love I Also, I love the fact that they
are like what you taught, Like Taylor, I think we're
very aligned here. Like this open source, you know, indie
tech thing is really important because I think ironically, like
what big tech actually does is it slows the development
of technology, Like it's in a lot of ways it's
anti innovation. And I think I wrote this like line
(26:01):
somewhere maybe when I was at the post about Apple,
which is like it's a lot more profitable to prevent
innovation than to innovate. And I really think that's true.
So I mean, if you're really anti technology, you should
be pro big tech.
Speaker 3 (26:14):
Ironically, well, I love that they that they prohibited like
taking photos of it, because I feel like, no matter what, usually,
you know, there's a stage moment so that you can
go viral on you know, social media, of course, And
I'm so happy to hear about this tailor, like I
hadn't heard about Summer of Flood before. And I feel
(26:36):
like this critique that separates, you know, this more nuanced
approach where you can love technology and hate the business model,
I think is is just a like a delineation that
will be so important coming forward because like one of
the biggest pushbacks against any kind of political action against
(26:57):
concentration of power in like the hands of a UAI
companies is well, people are using it, people love to
use it. Like, yes, I also I love to use it,
but you know you can they don't have the monopoly
on the technology. They are not the sole inventors or
the only people who can push this stuff. And if
young people are getting involved in thinking about open source
(27:19):
models for it. I think that's awesome.
Speaker 4 (27:21):
I just want to see them exert more political power,
and I wish that these kids, you know, it's frustrating
because to read the media coverage of a lot of
this stuff, it's like a reporter shows up for one
day they're.
Speaker 2 (27:33):
Like, Wow, these kids hate tech, they like the flip phones.
Speaker 4 (27:36):
Like it's just this very surface level coverage that doesn't
really interrogate their ideology or what they're trying to achieve,
and it doesn't mention that they are actually opposed to,
extremely opposed to this entire online safety slop about protect
the children that claims to be anti big tech, Like
these these kids are very anti surveillance. And part of
the reason they didn't want it all, like document it
(27:58):
was because they feel like they've lived their entire entire
lives where like everything is packaged and documented and shared
and surveilled, and so yeah, I just like, I don't know,
I'm heartened some of them are gonna apparently protest in
DC soon. I just want them to, Like, I want
these people to like get quoted more, you know, instead
of these like fake groups like design It for Us
(28:19):
that are just fake gen z groups that just like
pretend to be anti big tech but exists to push
metas policies.
Speaker 1 (28:25):
But it's sort of hard to quote them because many
of them didn't reveal their identities, right, I mean talk
about who are the organizers and do you think it's
strategic benefit or strategic mistake not to allow it? I mean,
I remember the Occupy wool Street days, Right, there's like.
Speaker 4 (28:38):
They reveal their idea. They're not They're not like masked
or secretive or anything. I mean you could talk to
them the during the actual event. They did have a
press puppet, so they had this puppet that would speak
for them to the media just because they didn't. It
was like this massive decentralized event and they didn't want
media just like going around like talking to random people
(29:00):
instead of like hearing sort of from them, like the organizers.
But they're not like hard to find. I mean I
spoke to these people. You could show up four or
four Media was there, Wired was there, The Economists was
allegedly there, so they're you know, again, this is a
massive public event in New York City with hundreds, if
not thousands of people participating.
Speaker 1 (29:20):
Taylor, do you have any sense of where this where
this goes this this movement, I mean, does does this
ladder up to be like a policy platform for the Democrats?
Speaker 4 (29:28):
For not for the Democrats. They don't like the Democratic Party.
These kids, you know, their mom Donnie supporters, they're veryed
to the left. I think they're part of this faction
of the left that's that's exerting pressure on the Democrats,
you know, and you see the Democrats struggling. This is
part of why they want to control the internet so badly.
It is because the establishment democrat sort of policies are
(29:49):
not popular among their base, and especially not with young
people like these who hate mass surveillance.
Speaker 1 (29:56):
When we come back read experiences that downside of using
AI agents stay with us, Welcome back.
Speaker 5 (30:15):
Read.
Speaker 1 (30:16):
We heard in the early part of this year about
token maxing. There's now token minning happening for Sudden eight
companies A being their employees to cut back on over
use of AI books of the cost of tokens. This
is something you, in a very human way, experience yourself.
Speaker 6 (30:33):
Yeah, well, while while you and I were in Can
hanging out, you know till the wee hours of the
morning or whatever, jet you know, my codex.
Speaker 1 (30:42):
My I hasten to add at a at a work
at a work adjacent to it was Creator Economy, slash Marketing.
Speaker 6 (30:49):
It was for sure, it was all at work, but
it was you know, while I was gone, I had
noticed a couple of bills for like five dollars in
my in my email. And you know, when you're in Europe,
I don't know, it's just I don't know about you, OZ,
but I always find it hard to like actually get online.
And then I was on vacation, you know, in the
(31:09):
Alps with the family and everything.
Speaker 5 (31:10):
It was great.
Speaker 6 (31:11):
But I come back and I'm like, it's like Sunday morning,
and I'm jet lagged, and I'm at my computer. I'm
trying to figure out what these bills are on my
you know, from my codex, which I hadn't been using.
Speaker 1 (31:21):
A codex read is is is open Ay's platform or
how do you define codex?
Speaker 5 (31:26):
Yeah?
Speaker 6 (31:26):
I mean, it's it's this Mac app that basically lets
you be pretend to be a software developer. It just
does everything for you and we'll you know, you can
basically start as many AI agents as you want and
have them work on these on these projects for you.
I love it. It's fantastic. Like we've I've built so many
cool things with it, Like I've gone from zero software
(31:47):
development to like using it all the time. But like
I hadn't set any limits on this thing, and it
I go to my credit card bill and it's just
all open AI bill, like for five dollars, just like
a huge It just goes on forever, and I'm like,
oh god, what so like this is partially me like
(32:07):
making fun of myself for the benefit of readers and
hopefully humanity.
Speaker 1 (32:12):
Well, what was the project that you would you were
trying to do and how did how did how did
open ai codex go rogue in terms of these repeated charges?
Speaker 6 (32:19):
Yeah, well what it What it was was that all
my projects got deleted and I had to reinstall the
software and then I but I before I deleted Codex
form my computer, I asked Codex to make a prompt
for me to restore my old projects once I reinstalled it,
because like the files are there, just like all the
chats are gone. But I had I had one of
(32:41):
these projects was particularly complicated and I won't go into
the details, but I tried to restore it right before
I left on this vacation, and I wasn't monitoring it,
and it somehow got stuck in this loop where it
was just reading all of this data over and over again,
like every day, and then running out of credits and
then like top up my my five dollars credits. So
(33:02):
like there were some days where there were like twenty
five dollars charges in a single day, and I just
had no idea this was happening for two weeks. It
turned out it wasn't that much. It was only like
five hundred bucks, I mean, which is still a lot,
but like it could have been thousands, you know, it
was it was like.
Speaker 1 (33:19):
Hey, you got and you've got lucky. What was the
five dollar how come it was in five dollars increment?
Speaker 5 (33:22):
It just tops itself up in five dollars increments? I
don't know.
Speaker 6 (33:25):
I like, I must have checked that box at some
point and totally forgot about it. But what was also
sort of funny about this is like, so to try
to figure out what had happened, of course, like I'm
going to ask Codex, right, So I'm like, I'm like,
can you figure out why you're billing me five dollars?
And it goes and does like an investigation and comes
back and it's like, here's here's what happened. And I'm like, okay, well,
(33:48):
then help me write an email to open AI to
get a refund, like I want to a refund for
my five rebox. So it's like, okay, here's I'll draft
this email. And here's like all the bill and everything.
It even went to my I had it go to
my credit card and like, you know, cross reference it
and make sure it all matched up. And then it's like,
(34:09):
but you shouldn't send an email.
Speaker 5 (34:10):
You should.
Speaker 6 (34:11):
You should send that basically their their customer service chatbot.
You should start a chat with it. So I started
chat and which of course is just an Opening Eye chatbot.
So now the chatbot is responding to me, and I'm
copying that response and pasting it into Codex, and then
Codex is responding and I'm pasting it into the chat
and I'm like, I'm just a go between between one
(34:33):
of Open a Eye's products and another Opening Eye product.
Speaker 5 (34:36):
Now like I'm the It was just hilarious.
Speaker 1 (34:38):
When it coast you is five dollars everything, right, But I.
Speaker 5 (34:41):
Got a nice little story out of it.
Speaker 4 (34:42):
Read this is this, this sounds like a nightmare scenario.
Speaker 6 (34:47):
It was a It was a total nightmare scenario, but
like also fascinating and and and still other than telling
Opening Eyes pr you know, one of their what what
did those people call it?
Speaker 5 (34:59):
The puppet? The media puppet or the media puppet.
Speaker 6 (35:03):
About this, like I never interacted with the human Like
I was the only human involved in this whole, Like
I still have not heard from like a Open. I
don't even know if Opening has human customer service reps.
So it was, you know, it was it was a
learning experience for me. And also just like I thought
it was relevant because we're having this big discussion now
(35:23):
in one of the big discussions in AI is the
cost of all this stuff, and companies are trying to
figure this out, and I'm like this must be happening
just constantly every day at these companies, Like people are
just on purpose or not on purpose, like putting these
things into these loops and wasting a bunch of just
you know, and it's it's a point in time, like
I think this stuff will all get solved, But I'm
(35:46):
also kind of like somebody's got to figure this out
for consumers. It's just not it's not a consumer product yet,
and I really think in the next like year or two,
there's going to be some company that just comes out
of nowhere and makes like go for ai that's just
like the ultimate consumer product, and everyone's like, holy shit,
this is it, and I don't. I mean, I used
(36:07):
to think it would be Opening Eye, it would be
Google or somebody, but I'm starting.
Speaker 4 (36:11):
To think, well, else, it's in the market. I mean,
they should be building now, we would know about it.
Speaker 6 (36:16):
Someone's working in stealth mode or something. But there's a
there's a Larry and Serge who are probably like just
writing their application to Stanford, who are going to create
this product?
Speaker 5 (36:26):
Is my guess.
Speaker 1 (36:28):
Natascha Taita, what does Read's Quoti retail make you think?
Speaker 4 (36:31):
I'm very parent, I'm very into like cloud code and
I try. I like made agents and stuff, and I
have little automated things. But I am so paranoid about
exactly what we described because I feel like I hear
from friends about like similar experiences. I just think it's
really easy to like spend money and not realize it
if you're not super adept at these tools, which I
(36:53):
certainly am not, and most consumers are not so yeah,
I do think that this is a problem that the
companies are going to have to solve. Ultimately, there's just
gonna be a lot of angry customers, if you know,
especially when you're starting to use chattupet for shopping, which
I do all the time now, Like I'm very careful
to go to like third party services. But you know,
I just readd my living room and I was taking
(37:16):
pictures of products and saying, hey, you know, chattupt find
me dupes. Basically I can't, you know, cheaper versions of everything.
But there's a world where I could just say, like,
here's a picture, find cheaper versions of everything, go off
and buy it and order it. I would actually love that.
That would save me so much time. But without safeguards
in place, like I would be so nervous. It'd probably
ordered me five hundred chairs, and yeah, I'd broke.
Speaker 3 (37:38):
For some reason, it just made me think. I feel
like there was some kid who racked up a massive
candy crush bill, like without his parents knowing, and there
must be so many of those happening right now, except
the cap is like insanely high. I mean, yeah, I
think like the lack of fine brain controls. Like even
just hearing that it went to your credit card read
(38:00):
is just like you know, because you hear about like
you hear about things going awry all the time. But
now I've got to go find the kids who racked
up like a million dollar token max bill. Maybe Larry
and Surge the future Larry and Sergey are actually just
in debt.
Speaker 5 (38:16):
Now that would be a great story.
Speaker 6 (38:18):
I mean, but yeah, I think it's like, but I'm
also kind of like, yeah, but what's the alternative? The
alternative is like you as a tech reporter, like you
just don't use this stuff, and like you're not you know,
you're not You're not seeing it. Like if you don't
I think, if you don't use it, and you're not.
Speaker 1 (38:32):
Able to expense the five hundred bucks the sem before
read in service of your general.
Speaker 5 (38:37):
Expensing the five hundred bucks because I wrote about it.
Now it's for a story.
Speaker 1 (38:41):
No, I'm just curious, what where does this? I mean,
just as we close, like, where does this fit in
the in the larger kind of token maxing to token minning?
Speaker 5 (38:49):
Ceasaw?
Speaker 1 (38:50):
I mean, there was a a New York Times story
recently with the headline tech workers maxed out their AI use.
Now they're trying to minimize it. And it's basically about
perverse corporate rewards and how employees were like basically promoted
on the basis of how many tokens they've burned. And
then the tech companies realize this is an extremely bad idea,
And now tech companies like Meta are trying to encourage
employees to use their internal models and not always use
(39:12):
the most advanced available you know models, because that max
is up too many tokens. Uber I think burnt through
all of its AI budget in the first four months
of this year. Like how does your experience kind of
map to the larger discourse about token token maxing? And
and I guess finally, is this a like a headwind
for open ai? And you know, Anthropic and Google in
(39:35):
terms of people come much more posimonious with wanting to
use this stuff.
Speaker 5 (39:39):
Yeah.
Speaker 6 (39:40):
I mean, so there's two there's two questions there, and
I think one is like there's a there's a pattern
in all of this where a new a new capability
or new product comes out and then it's very inefficient,
it's very expensive, and then it and then it quickly
you know, becomes about efficiency, right, and I think we
had the new thing here was was the harnesses the
(40:02):
ability to use like way more tokens, which started sort
of at the end of last year, and people got
very excited about it. Companies set up these leader boards
that you're talking about, and which of course is going
to just incentivize people to waste as much tokens as possible.
Speaker 5 (40:17):
And now we're already.
Speaker 6 (40:18):
At the point where the companies are like, okay, we
kind of understand this, now, now let's make it efficient.
So it's a very natural kind of up and down cycle.
And there will probably be something that happens, like at
the end of this year. You know that that causes
the same thing, the token usage will spike again. And
I don't know what that is, but maybe Natasha knows.
Speaker 3 (40:37):
But like, no, I think that. I mean, first of all,
everyone told them the leader boards were a bad idea
when they did them, Like this was not like a say, and.
Speaker 5 (40:46):
They knew it. They knew it, they knew that would happen.
Speaker 3 (40:48):
But I'll just say I think that this is I'm
seeing at least more debate around like you know read
you've called it like the token economy. I think I've
seen more debate around whether that is the way to
charge for these chatbots. I mean you even see like
Alex Karp, the CEO of Palenteer, talking about like why
would you pay for that when they are going to
(41:10):
you know, why would you pay for using this intelligence
when they can then take your data and you know,
like replicate what your company does or you know, if
you're a biotech like take that kind of information. And
I think that's really interesting, like pushing for a different
kind of payment or business model.
Speaker 6 (41:25):
Yeah, I mean there's also something happening here, which is
like there's what companies use internally with their software development teams,
and then what the wider teams use, and then and
then what they use for their customer facing products, and
of course the software development teams, I think they still
have all the best models. They're they're not skimping. Nobody
(41:46):
that I talked to is like skimping on that really,
and then on the customer facing stuff, Like there was
a CTO of like a major consumer brand who told
me that, you know, they are using all these Chinese models,
but they don't even want to tell the world that
because they'll get a letter from the commerce department. So
it's just I think it's like this is not like
(42:08):
one blanket thing. There's a lot of nuance happening there.
So I think, you know, Alex Krp's point was was right.
It's just that, like the the market is so large
that I don't think it matters for these companies like Anthropic,
like the the they have only started to really scratch
the surface in terms of like penetration, so there's gonna
(42:30):
be the pie is growing so big that they're fine. However,
I do think this is a headwind for like the
larger the company. You're more at a disadvantage here because
a small startup can come in and they can they
don't this doesn't matter, Like these these startups in Silicon
Valley will use as much as many tokens as they
want of the best models because they're they're first of all,
(42:52):
they're venture backed. Like it doesn't you know, they're not
like looking at profitability at this point. They're looking at growth,
and so they're just gonna max out on the tokens.
And a big company with forty fifty thousand employees or
more like they can't do that. The costs just go
so they're ultimately like they're more susceptible for disruption, and
I think you're going to see a lot of that.
(43:12):
Like the regulated industries, maybe that happens more slowly, but
in non regulated spaces it's going to happen fast.
Speaker 1 (43:20):
That's all we have time for today. Thank you all
so much for joining us.
Speaker 2 (43:23):
Thank you guys, thanks for having.
Speaker 1 (43:25):
Us for tech stuff. I'm mos Voloshin. This episode was
produced by Eliza Dennis and Sina Ozaki. It was executive
produced by me and Julia Nutta for Kaleidoscope and Katrina
Novel for iHeart Podcasts. Our engineer today was Eli Bronstein.
(43:48):
Jack Insley makes this episode and Kyle Murdoch wrote our
theme song. A special thank you to Taylor Lorenz, Natasha
Tiku and read Albergotti. Please check out all the work
they put out into the world. We're lucky to call
them friends at the pod
TechStuff News
Hosts And Creators
Oz Woloshyn
Karah Preiss
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
Betrayal Weekly
Betrayal Weekly is back for a new season. Every Thursday, Betrayal Weekly shares first-hand accounts of broken trust, shocking deceptions, and the trail of destruction they leave behind. Hosted by Andrea Gunning, this weekly ongoing series digs into real-life stories of betrayal and the aftermath. From stories of double lives to dark discoveries, these are cautionary tales and accounts of resilience against all odds. From the producers of the critically acclaimed Betrayal series, Betrayal Weekly drops new episodes every Thursday. If you would like to share your story, you can reach out to the Betrayal Team by emailing them at betrayalpod@gmail.com and follow us on Instagram at @betrayalpod and @glasspodcasts. Please join our Substack for additional exclusive content, curated book recommendations, and community discussions. Sign up FREE by clicking this link Beyond Betrayal Substack. Join our community dedicated to truth, resilience, and healing. Your voice matters! Be a part of our Betrayal journey on Substack.