July 24, 2026 • 43 mins
How much can we actually trust AI to stay inside the lines? Nitasha Tiku (The Washington Post) on the OpenAI models that went rogue — and why this isn't necessarily a sign of the end times… just sloppy security. Reed Albergotti (Semafor) on the company that wants to help anyone make software to power robots. Do we even have to talk about what that means for the future of human labor? Finally, Taylor Lorenz (User Mag) on Phoebe Gates' AI shopping startup Phia. Is it too early to call it nepo slop?
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 (00:12):
Natasha read Taylor.
Speaker 2 (00:14):
I'm not sure how much you think about students using
AI to cheat, or at least to get an edge
in the classroom, but there was some interesting live research
that emerged recently from Brown University. An economics professor made
his students take a final exam in person after suspecting
rampant cheating on the take home midterm. I want to
(00:35):
ask you to guess the delta between the average score
on the midterm and the average score on the final.
Speaker 3 (00:41):
Well, I saw this story, and it was a massive delta.
Speaker 4 (00:44):
Me too, Me too.
Speaker 5 (00:45):
I gotta say fifty percent.
Speaker 4 (00:47):
I didn't see it.
Speaker 2 (00:47):
The average score on the midterm was ninety six percent,
the average score on the final forty eight point six percent.
Speaker 3 (00:54):
But there was that one kid who like, actually did
he like got an F and then it went to
like a D or something like that.
Speaker 1 (01:05):
Welcome to Tech stuff.
Speaker 2 (01:06):
I'ma's Volosian and this is the Week in Tech where
I'm joined by three of the world's most clubbed and
reporters to break down what's really happening in tech right now. Today,
we're joined by reed Albrigotti, Take editor a Semaphore, Taylor
Lorenz of user mag and Natasha Tiku, tech reporter at
The Washington Post's welcome all.
Speaker 4 (01:24):
Hey everyone, thanks for having us.
Speaker 3 (01:26):
Good to be here.
Speaker 1 (01:28):
So Natasha.
Speaker 2 (01:29):
Nick Thompson, the former editor of Wired and current CEO
of The Atlantic, came on tech Stuff in early January
this year to give his predictions for the year, and
one of them, the top one, was that this will
be the year of the first legit AI catastrophe. Do
you think this was the week when it happened?
Speaker 5 (01:47):
No, I don't, But I think this is one of
many weeks in which we got a pretty clear picture
of what the next few years are going to look like,
which is messy, disastrous. Everybody taking incidents to try to
support their own narrative about where AI capabilities are headed,
(02:08):
and you know, how to protect ourselves.
Speaker 1 (02:11):
So what actually happened?
Speaker 5 (02:12):
So I mean this this is a really good one
to dig into. I think because last week Hugging Face,
which is kind of like get hub but for AI,
like they have data sets, models, it's very open source,
funded by all the usual vcs. They put up this
blog post that they had this unprecedented security incident that
for the first time, an AI agent had broken into
(02:36):
their system and tried all of these methods to get
access to some data sets, and it did it by
uploading like a malicious data set that would allow it
to kind of like get into the system through their
data processing pipeline and then.
Speaker 1 (02:50):
Like a trojan horses.
Speaker 5 (02:51):
It were yeah, yeah, And at the time, Hugging Face said,
you know, we think that this looks like an agent
that's used for security and RESO, but we don't know
what LLLM it is. We don't know what model. So people,
you know, obviously this is coming in the middle of
all of this talk about Chinese open source models, how
they're proliferating, how they're getting more capable, so nobody knows,
(03:13):
you know, what's happening. Then this week on Tuesday, open
Ai puts out a blog post and they're like, oh, hey,
that was us, and they said, you know, they discovered
after the fact that while they were testing both GPT
five point six Soul and some new unreleased model, they
were testing it on its ability to do these like
(03:35):
cyber offensive attacks, and they gave it this benchmark called
exploit benchmark that's supposed to test its ability to like
break into, you know, to hack into systems. The AI agent,
which was allowed to work you know, autonomously for a while,
decided to cheat on the test and figured a way
(03:56):
out of the sandbox that open ai created, like based
you know, testing environment, got into the Internet, decided to
break into hugging Face. And open ai is so lucky.
They are so lucky that it was another AI company
that has invested in in the industry and the narrative
because this is I mean, even the the CEO of
(04:19):
hugging Face was like, yeah, don't do this to us.
It's illegal. You know, they had notified law enforcement. And
now the announcement that it was open ai came as
part of a partnership between hugging Face and open Ai,
which was obviously you know, put together after the fact.
Speaker 2 (04:34):
So the model had essentially determined that it would be
easier to go and steal the answer from hugging Face
than to come up with it itself, and therefore did that.
Speaker 5 (04:44):
Yeah, it knew that it was being tested on a benchmark.
I think they even maybe you know, uploaded the benchmark
from hugging Face and it figured out it was supposed
to not be able to access the Internet.
Speaker 2 (04:58):
But that's a little I mean that That's the part
why Zone zoned in on the word catastrophe, because the
bit where is not supposed to be able to access
the Internet where where, but it does. I mean that's
got a kind of cinematically spooky quality.
Speaker 4 (05:10):
Well, we failed to secure the sandbox, right, Yeah, exactly.
Speaker 2 (05:13):
Failed to secure the soundbox. Premagine preschool is hearing that
in the nineties.
Speaker 3 (05:18):
What are you doing?
Speaker 5 (05:22):
Yeah, you zeroed in on exactly the right part of
the confusion around this and why people are using it
to like support you know, various narratives. You know, people
who are very worried about AI super intelligence and AI
capabilities growing really fast. You know, we're talking about this
as like this is what we've been warning you about.
(05:44):
It's a rogue AI. You know, we gave it a
simple set of instructions and it decided to do it
in a way that you know, was not was not
what its developers would want. But many other cybersecurity professionals
were like, why did you not have a properly configured sandbox,
Like you could be using an air gap, which would
(06:05):
not have allowed it to break into an Internet system.
They allowed it to use like package installs, and I
think just a lot of cybersecurity professionals found this like
very sloppy. This is not you know, so it's not
it's not as though, oh my god, we can't contain
AI and it's doing some you know, like you know,
people were throwing around the word sentient machines and that's
(06:26):
not what this is.
Speaker 3 (06:27):
Well, on top of that that the model had no safeguards, right,
there were no you know, it was a really raw,
you know model, So.
Speaker 5 (06:35):
Yeah, they were testing its ability to to do these
kind of exploits, so they didn't have any safeguards on.
But then they just didn't have like traditional cybersecurity controls
that you would have in place when you're testing. And
this is something people had been warning against that maybe
the biggest danger is not when your systems are in deployment,
because then you have all your safeguards up. It's refusing
(06:57):
a lot of requests, but it's actually during this kind
of testing environment. And I'll just say one more thing.
This incident followed another blog post last week from open
ai about how they had to stop even just testing
an internal model because it like they weren't monitoring it
and it started doing all of these things because they
(07:18):
let it go on and on, because that's you know,
that's how these models get more functional. They're able to
like think and try thousands and thousands of different ways
of breaking into things.
Speaker 3 (07:27):
Well, I think that I like agree with you, Natasha
that it's not this isn't like some you know, code
read catastrophe at all. But I do think there's something
here that is sort of like you know this, The
AI security people do have a point on this. It's
a point they've been making for a long time, which
is that like, when you tell an AI model to
(07:48):
do something, it will, like the smarter they are, the
more they're just going to find some shortcut to do that. Right,
this is the whole argument. But this is the argument
with like the you know, the paper clip argument, Right,
if you tell AI to make paper clips, it will
just ultimately like find the best way to do that,
even if that means like turning humans into paper clips.
Speaker 4 (08:07):
So but also but also read, I think that that
is a is a sort of ridiculous concept because any
AI that's smart enough in that way would also have
the reasoning to know, like they're not going to turn
the world into paper clips. That's not the goal. Like,
I don't know, I think a lot of this like
kind of like what Natasha was saying before is like
a little it's being taken by these people. It's like
(08:29):
they made a sloppy security environment, the AI went and
solved the problem as it should have or whatever, And
and that's not evidence of it being super sentient or
even this being a particularly good AI agent. Mostly it's
just that like they didn't set up this test very responsibly.
Speaker 3 (08:47):
Definitely agree, I agree with but again, like I'm not
disagreeing with you at all. I just think there is
one sliver of this that is worth sort of looking at,
which is like, you know, they do sort of find
these shortcuts, so and in this case, the shortcut meant
like breaking the law and hacking into hugging face, so
I and I think the other thing to remember is
that like we still don't know how these models work,
(09:10):
Like no one has been able to figure out like
what are these neurons inside these models actually doing? And
I do I think the big question is like, as
they get more and more capable and more powerful, is
it is it enough to just put you know, the
safeguards in place, which again we're not like they didn't
what I'm curious about because this this this Natasha, you
(09:32):
mentioned there were two open Ami models of working on this.
One was a publicly available one and one has not
yet been released. So could could I Well, probably not me,
but like could somebody has a bit more sophisticated than
me use open Ami to hugging face exactly like this
went down like using the publicly available models. Or was
there something special about this unreleased model? Well, no, because
(09:53):
there were no safeguards like the publicly released one will
have will have controls on it that will not let
you do this unless you can jail break it, which
you know is getting more difficult to do, right.
Speaker 5 (10:04):
Okay, I will say, I mean, I do think it's
important to note that this is like many people are
interpreting this as like we were right about the paper
clip maximizer. You know, this is happening in this way.
But I think there's a way that because they conceived
of it, of the problem this way, they approached security
in a certain way, like you could have been thinking
like a cybersecurity professional the whole time and not thinking
(10:27):
about AI alignment, aligning it with with human values. And
one of the things I mentioned that like prior open
ai blog post. What they realized is that they didn't
have sufficient monitoring systems. When you let a system go
on and on for a really long time, they weren't
watching it closely, like there are simple mechanisms, you know.
Speaker 4 (10:49):
It doesn't even go.
Speaker 5 (10:50):
Back to what Reid said about knowing what's happening on
the neurons, Like they weren't even watching what it was
doing during the security test in a secure way.
Speaker 2 (10:59):
Ax out there saying that saying maybe open Aye wanted
to demonstrate anything mythos can do, we can do better,
or is that it's great marketing for opening I.
Speaker 5 (11:08):
For sure, I think the way that they wrote the
blog post at the top, it says like we are
treating this as an unprecedented cybersecurity incident that shows the capabilities.
And I will say, like, I don't think anyone's arguing
that these lms make people much much much better at
hacking and that it's capable of improving and enhancing and trying,
(11:31):
you know, cyber offensives that humans can't do. But everyone
I talked to was like, this is not an example
of enhanced capabilities. This is an example of like kind
of a sloppy testing environment.
Speaker 3 (11:42):
I mean, we all agree on that. I think it's
just it just sort of highlights this question about the future,
which is like, when these models become more powerful, do
we need do we actually have to understand how they work?
Speaker 2 (11:52):
Has this has this effected the kind of policy discussions
in Washington? What's how's the White House and crapsios and
others around around Trump and the sort of technology advisory
space reacted to this? Is this is this like grist
to the mill of the of the deceleration lists or
is it?
Speaker 5 (12:09):
I would say they were very busy with They're still
very busy with Chinese open source models.
Speaker 4 (12:14):
So that was the focus.
Speaker 5 (12:15):
This didn't I don't think it got the attention of
you know, the the same White House folks they were
putting out. They were putting out information about distillation, which
we talked about last week, and you know, did did
Chinese open source models steal from quote unquote steal from
from anthropic? But there were a number of regulators that said, like,
(12:39):
it's time to legislate, it's time to get involved, Bernie
Sanders and many others. So yeah, it's feeding into this
growing sense of urgency around having some safeguards in place,
But I just want to say, like another plea for
people to start thinking about monitoring like downstream usage, like
(13:01):
kind of what reads saying about the neurons. Actually, now
that I go back to it, because what they do
when they try to figure out the models like reasoning
is look at its chain of thought, like look at
this little like they give the model a scratch pad
and it says like, ah, it might be faster to
go to directly to hooking face, let me break in
or whatever, and we don't even know if those that
(13:23):
text actually reflects its interior thought. You know, it's like
decision making process. So it's true, this.
Speaker 2 (13:31):
Is the dome worry mom and dad, I'm doing my
homework upstairs. Basically, of AI, it's thinking out loud.
Speaker 3 (13:37):
It's thinking out loud, and that is actually this is
now this, this is actually the state of the art
in like AI safety is like looking at it's thinking
out loud.
Speaker 2 (13:45):
But but it can think out loud deceptively potentially yes.
Speaker 3 (13:48):
And also you're not looking at it's thinking inside. It's
like if you if you only base what somebody's thoughts
are and what they say, then that's you're clearly not
getting all their thoughts right and open.
Speaker 5 (13:59):
I wasn't even reading that you know, so so.
Speaker 3 (14:02):
Right for this experiment.
Speaker 5 (14:04):
They were let's let's yeah, let's let's clean it up.
Speaker 2 (14:08):
So here's my favorite detail from this whole story, which
actually comes from Forbes, of all places. Quote Hugging Face
said it had first attempted to use an undisclosed AI
model from leading US Labs to defend against the attacking
AI agent, but the guardrails around that model cyber capabilities
stemied its response teams work. The company said it instead
wound up using an open source AI model from Chinese
(14:30):
company Zai to carry out its defense.
Speaker 4 (14:33):
Which is I think exactly why these type of safeguards
that we're sort of putting in our American models are
deranged and we need access to open source Chinese models.
I mean, I think even just from a security standpoint.
Speaker 5 (14:48):
Right, I mean, many cybersecurity people were saying, you know,
this shows the need to like have policies that help
proliferation of open source models, because as like, history has
shown that cyber defense capabilities rely on open source models,
So it doesn't need to be the Chinese if we
had been investing in open source AI or if our
(15:10):
leading companies were putting out open source AI models, it
could be us. But they did talk about the need
to make sure like hospitals are adopting the latest open
source AI models in order to make sure they're capable
of defending if such a thing happens to them.
Speaker 2 (15:26):
Okay, lightning round, before we go to the break, what's
the one thing from each of you that somebody's listening
to this podcast needs to know to sound smart.
Speaker 1 (15:33):
At the weekend?
Speaker 4 (15:34):
About Kimmy K three, I mean, I think everybody wants
to act like, oh, it was distilled. It was definitely distilled.
I mean, this is a model that I believe was
testing and also if I might be getting this wrong,
but I also think that it's this model also has
capabilities that our American models don't have. Or it is.
Speaker 1 (15:53):
Kimmykre right, Yeah? Which is what?
Speaker 2 (15:55):
Which is a new model from a Chinese lab that's
kind of outperform throwed on various benchmarks and therefore everyone
kind of lost their mind?
Speaker 1 (16:03):
Or what's the what's the like?
Speaker 3 (16:04):
It's sort of on par with with the with the Frontier.
I wouldn't say it's outperforming, and I don't know if
I do, you mean it has capabilities the US models
don't have, or they use training techniques that are sort
of novel and that you know, the US labs could
learn from.
Speaker 4 (16:19):
Basically, Like what I was reading yesterday, I was saying
that like the architecture of the model is superior to
to whatever it was claiming to be distilled on, and
that these are not developments that they could have stolen
basically because it is they have developed further on you
know whatever. I guess it's like the latest sort of
Claude model. So I just I think this idea that like, oh,
(16:42):
these are all just like t MoU versions of Claude.
I don't think that that's a full picture. Yes, I imagine
that they're doing some level of distilling. They all do. Like,
I just think this is like an industry.
Speaker 3 (16:55):
No, there's an innovation there. Like I think what you're
saying is like you can distill from one of these
large models in order to shortcut the training, but there's
still innovation in the training in being just because by
virtue of being hamstrung on the amount of compute they have,
they have to they have to take these shortcuts and
they have to find new and novel ways to train
(17:17):
with less compute, and so there's innovation there for sure.
Speaker 2 (17:25):
When we come back, read takes us to a garage
full of automated pickup trucks carrying anti drone weaponry.
Speaker 1 (17:48):
Welcome back.
Speaker 2 (17:49):
So I mentioned in the first story that Nick Thompson
predicted this would be the year of the first legit
ai catastrophe. Our panelisting that that that prediction has not
yet been proven out.
Speaker 1 (18:00):
My prediction was was the year of the robot Read.
Speaker 2 (18:03):
You had a story this week about a company that,
according to the headline semiphore quote, wants to turn robotics
into child's play.
Speaker 1 (18:11):
What's the company and what's it do today? And what
does turning robotics into child's playing?
Speaker 3 (18:16):
Yeah, so the companies applied intuition and I went visited
their their headquarters on Monday and took a tour through.
You know, they have this garage where there's really just
a bunch of cars in you know, various states of disrepair,
so like you just see all the electronics around a car. Essentially,
what they've been doing is working for a lot of automakers,
(18:39):
most of the big automakers actually and building the technology
for them everything from like the software platform, infotainment systems,
et cetera, to the self driving tech and they've been
making a lot of money doing that. They've now expanded
into like farming and other you know, construction, other areas.
Speaker 2 (18:56):
So basically, if you're not Tesla, your license applied intuition
evan technology.
Speaker 3 (19:01):
Yeah, exactly. Like if you're an automaker and you realize
you don't have the tech talent that Tesla has, but
you want to have a Tesla like product, you kind
of you hire this this company rather than try to
do it in house. But and that's been a good
business for them. But what the news was on Monday,
I wrote I wrote about it on Tuesday morning was
that they've come out with this new open platform that
(19:24):
you can you know, it's a product that you can
pay for and you can actually use all of their
data and all the infrastructure they built over almost a
decade and train your own robots. Essentially. They would call
it physical intelligence, which is this jargon term for like
any machine that can be programmed essentially. But like their
example of is, like you have a kid who wants
(19:46):
to make an autonomous lawnmower, they could do it now,
whereas it used to take a team of like eight,
you know, PhD engineers to do all the computer vision
and train models, et cetera. And I played around with it.
I mean, it's still I would say, this is not
child's play yet, but it's moving in that direction. And
I just think it's really interesting because it's like, I
(20:07):
don't know about you guys. I play around with like
Raspberry pies with my with my ten year old, you know,
and we try to hack stuff together, and I'm like, oh,
this is cool. Like this allows to do even more
fun projects. And then I think sort of create more
entrepreneurship around AI for the physical world.
Speaker 2 (20:23):
Okay, but so I could go and buy a regular
NORM from home depot and then use this platform to
turn into a robot.
Speaker 3 (20:31):
Yeah, well you could put like a GoPro on it
or something, right, and go film film yourself mowing your
lawn and then upload the video into this platform and say, okay,
I want you to basically do like an evaluation and
train an AI model that is the perfect model for
mowing my lawn, you know, and be done in a day.
Speaker 5 (20:51):
Essentially, you provide the hardware and they provide the intellig This.
Speaker 3 (20:57):
Is just a software platform. There's no they don't do
they don't do any hardware, so you make the hardware
well or like it's really kind of right now aimed
at like startups, right like if you were it would
be like a more like a company probably trying to
build a n autonomous lawnmower at this point. Like it's not
at the point where you know, but.
Speaker 5 (21:16):
But it's only for hardware, Like it's not. It's not
like you wouldn't use this just as an online model.
Speaker 3 (21:23):
It's just a software platform that you can train like
computer vision models. Look, I'll tell you what I want
to do with it. Do you want? This is embarrassing,
but yes, but I didn't put I didn't put this
in the article because I was worried about animal rights people.
But I'm going to say it anyway. I have a pool.
Speaker 4 (21:40):
I have a pal to a vegan on this call.
Speaker 3 (21:44):
I have a pool. And no, it won't involve eating
any animals. I mean I could go there, but but
I have this pool. And the ducks in my neighbor
they fly into the pool and they just make a
huge mess. They just poop all over the pool, cover
all over everything. Right, And I'm always trying to get
the ducks to stay away, and there's YouTube videos like
(22:05):
I'm not the only one who has this problem. I
just want to take a little Raspberry Pie, program it
to roll around and just scare away the ducks. Maybe
have some sort of like NERF gun attachment that shoots
a little NERF gun. Is that bad?
Speaker 5 (22:19):
Like?
Speaker 3 (22:19):
Am I a bad person for wanting to do that?
That's the product I want to build.
Speaker 4 (22:22):
I think that's honestly amazing. And you're not going to
hurt the ducks, And I don't want to hear that
I just want to scam. You're not trying to build
an autonomous weapon. I think. I think that sounds great.
Speaker 5 (22:35):
Isn't this the Tony Soprano problem?
Speaker 1 (22:39):
He liked it. I think he liked the ducks. I
think he Yeah, I.
Speaker 5 (22:44):
Support this, Yeah, but it also sounds like you're going
to have to do all the work. I don't think
that this that applied intuition is going.
Speaker 1 (22:51):
To help well.
Speaker 3 (22:52):
So I buy the g So, I buy the go Pro,
I mean sorry, I buy the I buy the Raspberry Pie,
and I find some little one of those little robe
I even I think I have most of the hardware
I need. I just I just need to upload the
video and essentially train a model to recognize ducks and
you know, fire a little a little nerf gun in
(23:12):
its general direction, not at it to hit it.
Speaker 2 (23:16):
Read for the for the for the layman. And I
put my hand up here, like what is the what
is the problem in robotics? This is solving like what what?
What isn't what was not possible yesterday that maybe possible
today because of this development.
Speaker 3 (23:30):
Well, I I don't think this actually is not it's
not about like this isn't like an advancement in AI
or robotics. This is just democratizing it. I know you
all hate that word, but it's but it's letting people
like me or startups you know, with with fewer resources
get into this field.
Speaker 2 (23:47):
To use natural language to to essentially train physical robots.
Speaker 3 (23:53):
Right exactly like I want to I want to scare
the ducks away? Can you create a model that will
recognize the ducks and you know, do whatever. So it's
like a move in that in the direction of like
you know, we'll have more than just the robot vacuums
in our house, like we'll have a and you know,
eventually humanoids, but that's farther down the line.
Speaker 2 (24:15):
You talked about something your story called the missing link
effect in robotics. What is that and how does that
kind of play into this story?
Speaker 3 (24:23):
Yeah? That was sort of like this this concept that I,
you know, you're essentially like you can create. There's robotics
everywhere now, right, Like you go into a biotech lab.
I was in Boston last week, right, there's there's robotic
pipe heading machines in all these labs. Now, there's there's
all sorts of automation that you know, centrifugas, et cetera,
(24:46):
that are testing DNA and doing all sorts of stuff.
But like, you still need a lot of people to
run these labs because they're still built for humans. So
it's it's not like it's not a step chain. The
robotics gives you these incremental gains and efficiency. But my
point was like, let's say, and this is what a
(25:06):
lot of lab technicians say, not all of them agree,
but like, let's say you could have a humanoid robot
that has like five fingers and can do you know,
any sort of dexterous tasks that a human can do,
and you can just program it. All of a sudden,
every single biolab as they exist today can become completely automated,
and then you can just have these things run twenty
(25:28):
four to seven and that would like allow scientists to
run more experiments that they would never do today that
just there just wouldn't be the resources to do today,
And so that becomes like an exponential gain in scientific discovery,
and that sort of principle of like you plug it once,
you get once these robots become more general purpose and
(25:49):
you can sort of plug them into you know, all
the technology we already have, you you get exponential gains
versus just Okay, now I got like a ten percent
increase in efficiency, you're twenty percent.
Speaker 2 (26:01):
So right now, for example, you could have a robot
that shot sort of nerf arrows at the ducks, but
then you have to go and pick them up yourself
and reload the robot, whereas in the future you could
also be a general purpose robot collect the arrows.
Speaker 3 (26:15):
Sure, yes, you do that, but then think about that
in like at an industrial scale, right. Building construction is
a great example, Like there's all this automation and construction, right,
but you still need a ton of people, and that
creates like there's labor shortages, that creates like the bottleneck essentially. Right,
But once you can fully automate things, the things move
(26:38):
just exponentially faster. And that's like to me, that probably
in my view, and you might you all might disagree
with this. I think that's where humanoid robots are valuable.
Like people make this argument that, well, why do you
need humanoids, Like you could just create a special robot
for each individual purpose, but it just it doesn't work
that way, Like.
Speaker 2 (26:58):
The industrial wood world was designed to be navigated by
real humans and therefore humanoid robots and are actually they
fit into the system we've built better than any other
types of robots.
Speaker 3 (27:10):
Yeah, totally. I think that's I think they may be
a necessary step to get to like a thick full
full automation and like you know, fully roboticizing you know,
the economy.
Speaker 5 (27:22):
But wait is the miss So the missing link is
the technology that will make humans obsolete? Is that?
Speaker 4 (27:27):
What?
Speaker 5 (27:28):
What is the missing link?
Speaker 3 (27:29):
I still I mean, yeah, that's the that's the and
and look that was I put that in the articles
down like that. You know. The obvious next question is like, Okay,
what does that mean? For human labor. I personally like,
first of all, this is this is a ways of way, right,
Like this is not gonna happen overnight. But second of all,
I think that you just think about the economic growth
(27:49):
that comes from that world, Like I don't. I'm not
worried personally that humans are just gonna have nothing to
do like that that just is not gonna happen. We'll
find stuff to do.
Speaker 2 (28:00):
You did refer on your story to the Nobel Prize
winging economist Darren Astimolgulu's research, who found that every industrial
robot added per thousand workers measurably reduces employment and wages
in the community where it lends.
Speaker 3 (28:13):
Yeah, I just disagree, though. I just think that you're
not You're not thinking through all the ancillary benefits of
being able to do this kind of stuff and then
the economic growth that will happen.
Speaker 2 (28:25):
I mean, but according according to according to the quote,
is measurably reduced.
Speaker 3 (28:30):
Right. But like economists, they can only measure what already exists, right,
And this is this is the mistake everybody. It's like
you almost can't win this argument because it's like who
would have thought, like in nineteen hundred that, you know,
all the jobs that exist will be gone. But like
people will make millions of dollars sitting in front of
a webcam and just talking like the idea of a
(28:54):
creator could not have existed in their minds, right, So
there's I think humans will always find something that they
do that they're that we assign value to and will
be just fine.
Speaker 4 (29:06):
Natasha Taylor, I'm just in my mind like right now,
daydreaming about what kind of robots I could have, you know,
automated parts of my life, because there's I'm thinking of
Reed's pool example, and I'm like, there's a few things
I'd like. I would like something to handle my garden honestly.
Speaker 3 (29:23):
To prune or what would you will, yeah.
Speaker 4 (29:25):
Prune weed, anything like water. I mean, they have automated
watering systems. I should just buy one off Amazon. But
but you know, I went out of town recently and
I needed to pay someone to come over and like
harvest the jilapenos that we're going to go rotten. Then
you know, just like manage it.
Speaker 5 (29:44):
I guess a jilapeno harvesting robot.
Speaker 3 (29:47):
Of course, I think a gardening robot is an even
more lucrative idea than than the Nerf nerve shooting duck robot.
Speaker 2 (29:54):
I would I have nightmas whenever I travel, I have
nightmas about the garden just kept becoming you know, wasteland,
and I feel like I would, I would be a
very irrational spender on more about the plantain my garden.
Speaker 4 (30:09):
Just a humanoid robot that lives in your backyard, that
can tend to your garden while you're.
Speaker 2 (30:13):
Gone, Natasha, I feel like I feel like you've got
to a counterpoint here.
Speaker 5 (30:20):
No, I mean I I have heard this argument about,
you know, we can't possibly comprehend the future like future
jobs from many, many, many executives. But I do think
it's really interesting the humanoid robot example in a lab.
Both my parents were like research scientists, and just thinking
(30:43):
about how it could potentially change that. I mean, first
of all, it's like just imagine the number of mistakes
and errors that could be made. But also, you know,
so like you can't you can't make the humans obsolete
because it's never going to be one hundred percent accuracy.
But in terms of all of the different things you
could try, Yeah, I mean I think that's that's like
(31:05):
pretty exciting when you apply it to like actual innovation
like new drugs, new science, new materials.
Speaker 3 (31:15):
Well, here's a here's a peta point for Taylor like that,
I think this is the thing that people aren't thinking
about when you have so when you have like fully
robotic labs, right, what is like the number one way
that we test all these molecules. It's animals.
Speaker 5 (31:32):
Oh, you know, people are building these like synthetic animals
so that we won't have to do testing and synthetic brains.
Speaker 3 (31:40):
That is true. I looked at synthetic brains under the microscope,
like they look like little brains. I mean that's just
brain tissue. These or they call them organoids. It's fascinating.
But like but right now, like so one of the
researchers said, well, if you look at like the C elegance,
which is like this little worm that that was made
huge advances and like the longevity industry like all this testing.
(32:02):
They like sort of root force tested the DNA of
C elegance in order to you know, to essentially like
do like do testing on it. And that led to
a lot of this a lot of this new research,
right and they're like, well imagine if like but that's
a very simple organize and imagine if you could do
the same thing with mice, and then we could have
(32:23):
this greater understanding of mice and that would advance medicine.
And I'm like, I'm picturing like like a robotic you know,
twenty four to seven robotic like warehouse just full of mice,
being like they die eventually right in these tests. And
I'm just like, I think this is a big animal
rights like once it gets to that, it's already controversial.
Speaker 2 (32:44):
Well, actually I want to share this in the in
the first segment, but but I didn't, but I will now.
So I had Stewart Russell on tech Stuff recently, who
was Elon's only expert witness in the open AI try
it about AI safety and one of the great sort
of advocate of the alignment problem and the paper clip
issue and all the other things. And his example was,
(33:05):
if you told the super intelligent AI to cure cancer,
the most rational first step would be to give find
a way to give all humans cancer, so you could
run as many simultaneous clinical trials as possible. So here
we are the connective tissue between segments one and two.
But now we go to the atbreak when we come back.
Bill Gates's daughter Phoebe gets into some startup trouble. Stay
(33:28):
with us, Welcome back, Taylor. I remember when Phoebe Gates
announced her AI shopping startup a little over a year ago,
investors from Kleiner Perkins to Sydney Sweeney piled into Fear.
(33:52):
Tell us what fear is and what's been going down?
Speaker 4 (33:56):
Yeah, Fear. It's Phia. I feel like it of it's
hard to pronounce. So it's an affiliate shopping site. Basically,
it promises to give you deals on you know, whatever
you're shopping on online. It's very similar to Honey. It's
essentially the same business as Honey, which is very funny
because Honey got into trouble for the same thing like
(34:19):
a year ago. I think. I guess what's different is
that it promised to be powered by AI, so you know,
AI powered shopping assistant, help you shop, help you find
the best deals, blah blah blah. Well, it turns out
that they were doing the exact same thing that Honey did,
which is sort of essentially stealing other people's affiliate revenue
and using it as their own.
Speaker 2 (34:40):
So how exacted as well? I do remember just last week,
you were talking about how useful it would be. Honest,
two weeks ago, I think how useful would be to
have an AI go and find lookalike products for what
you wanted for your living room.
Speaker 4 (34:53):
Yeah, this is not This is.
Speaker 2 (34:54):
Not the offering that that's solving that problem yet.
Speaker 4 (34:57):
No, Yeah, it's I think I believe it. It's a
browser extension. Basically, there's this messy world of like advertising,
cookies and data. You know, lots of data is being
harvested on us all day. There's lots of things that
are happening on our browser that were maybe not aware of.
So say you went to say you were a FIA user, right,
(35:18):
and then you went to buy something on Nike dot com.
FIA might get credit for that sale even though they
didn't directly drive that sale, And so that costs retailers
a lot of money.
Speaker 2 (35:27):
So it's not it doesn't it doesn't cause consumer harm.
But it's basically it's basically just stlling out the other
affiliate affiliate marketers and harming the retailers.
Speaker 4 (35:37):
Yeah, and the idea is that if this happens on
a broad enough scale, of course, retailers will raise their
prices to deal with this stuff.
Speaker 2 (35:45):
And you had Ben Adelman, who was a researcher looked
into this story on your podcast. So what did he
tell you and what made you want to kind of
cover this one?
Speaker 4 (35:55):
I mean, I'll be real with you, guys. I just
thought this story was interesting, not from any of this
stuff like il like it's a pretty standard. It sounds
like Honey already did this kind of scam. It's fraud,
it's not good. It sounds like everyone does it not
to make it whatever. What I found interesting is the narrative.
And I'm very skeptical of like sort of these takedown
pieces on female founders and everything. But this is a
(36:18):
girl that raised forty three million dollars, that has some
of the top investors, you know, that got in Forbes
thirty under thirty, that is just getting unfathomable amounts of
sort of opportunities and ushered into this like tech elite
world basically for building what was essentially a Honey clone
(36:41):
and ultimately doing the same thing that Honey was doing.
And so I just I guess to me it was
it was more interesting. It is like a story about
kind of like how does access play a role, how
does fame? How does adjacency to power like play a
role in this Silicon Valley start up funding ecosystem and
the companies that are sort of considered successes.
Speaker 3 (37:00):
I mean, if you could raise that much money, snap
your finger, well you probably could, honestly, Taylor, I mean
you're pretty well, but like you could start, you could
build a robotic you know, Gardner, Like, I mean, let's
do something more ambitious. I totally agree, Like what.
Speaker 4 (37:15):
Right bart headline? Taylor tries to put Gardners this no,
and it's like I don't want to like downplay like
the idea of like AI shopping, but I just I
feel like we can think bigger about technology and about
like what we're like lauding and what we're like praising
and putting founders on like most exciting startup founders lists.
(37:38):
Like you know, that's kind of to me. I'm like hmmm,
And you know a lot was focused on sort of
Phoebe Gates is like, you know, she's an influencer, she's
out there, you know, building an audience.
Speaker 5 (37:50):
And even when it launched, I thought it was like
a weird move for her. I'm like, you could do
you could have better nepotism than this you know, you
could have like a more impressive startup. She was talking
a lot about reproductive rights and like going in a
certain way with her influencer career. I thought this was
really a like lateral slash step down.
Speaker 3 (38:09):
I mean, it's like one of those businesses where you're
like you type into like CHATGBT, like find a way
for me to make money?
Speaker 5 (38:17):
Yeah, like ask your dad for help. If you haven't
been asking your dad for ask him for help. Because
this was not a good one.
Speaker 2 (38:25):
Okay, but let's so so tech tech NEPO babies obviously,
you know PB Gates is kind of intriguing as one
read jobs has his has his fund investing in like
cancer cancer technologies. I'm not sure if I'm not sure
if anyone knows much about that or where it's going.
But if you do, chime in, and who are the
other tech NEPO babies who want to be in tech?
Speaker 1 (38:48):
I guess my question.
Speaker 5 (38:49):
There's already a lot of them in tech. Tim Draper's
sons are investors. You would be surprised if you just
look at you know, like middle management at the fank companies.
I think you see a lot of a lot. Yeah,
so well, Sophie Schmidt, she had she had Rest of World,
(39:13):
and I'm not sure how long she's going to be
affiliated with that. I think Rest of World is actually
an awesome publication. So that's that's one of the technico
products that I would definitely stand behind. Yeah, I feel
like often they bring them in on their funds.
Speaker 4 (39:29):
Yeah, I mean I would say a really famous one
is David Ellison, Larry Ellison's son, who would be.
Speaker 1 (39:37):
That would probably the example number one.
Speaker 5 (39:44):
And we, like I have been thinking about this for
I guess decades, because we are not prepared for the
like the way that the that the children of these
tech billionaires are going to influence our world old with
the amount of capital that they will have access to
and the amount of like Taylor said, access just generally to.
(40:06):
I mean just imagine like Davos twenty third.
Speaker 3 (40:10):
What do you what do you think it'll look like?
Like we've seen that we have the Rockefellers, we have
you know, there's there's all these families, but like, is
this going to be just more of that effort?
Speaker 5 (40:21):
It's right, No, I don't think it's going to be
like the Rockefellers at all, because I think it's going
to be a lot of the kids trying to make
a name for themselves. So they pick a they pick
a cause, they pick a charity, they pick you know,
something that they want to be associated with. So the
money that would normally go to like kind of traditional
philanthropies which have their which have their own issues, is
(40:44):
probably going to go to like a startup, you know,
they want to when they show up at a Davo
circuit or whatever it is. They want to have like
something behind their name that's not just their mom or dad.
So yeah, it's just going to be I guess like neposlop.
Speaker 3 (41:03):
Some of them. Some of them might be good.
Speaker 4 (41:06):
Yeah, I mean I think it'll be interesting too, Like
with all of Elon Musk's kids, you know, Elon Musk,
like a lot of these Silicon Valley men also are
like obsessed with pro creating. They want to ensure that
their legacy continues, you know. So I wonder if all
said they're going to take a more like hands on
role than maybe some like finance guy or whatever, you know,
(41:26):
would have done with his children.
Speaker 5 (41:28):
And I mean so many of them got radicalized from
their children, right, Like their children's politics radicalize them. So
maybe they will be like the class traders that tried
to push more, you know, more income equality or different things.
I mean, I'm not saying it's necessarily going to be slopped,
but you have to imagine. I mean, their dads are
(41:50):
the ones who are like Elon's like I did it
by myself. You know, I'm an immigrant, I like didn't
have any help. And so these kids are going to
have like a lot of chips on their shoulders, I'm saying.
And the.
Speaker 3 (42:03):
Yeah, it's like dead right, Like they're not going to
give it away, are they? Is the philanthropy dead is
just going to be startups?
Speaker 4 (42:11):
Well, I would say maybe the Dario. You know, we're
about to get a lot of ea anthropic millionaires and
billionaires they famously kind of give to a lot of causes.
Speaker 5 (42:22):
Well, okay, but I went to this event in San
Francisco called what should we do about all this money?
Speaker 3 (42:32):
Get it out of California? Get the money out?
Speaker 5 (42:34):
Yeah, and I think they're going to go into donor
advice funds, which means like no transparency. There's just a
lot of ways to like set it up as though
it looks like a charity but there's no transparency, you know,
public accountability. And I think it's quite possible too that
a lot of this money will be going into AI safety.
Speaker 3 (42:52):
Oh god, well we'll have very safe AI. That'll be great.
Speaker 4 (42:57):
Yeah.
Speaker 2 (42:58):
Right, Well, that's all we have time for today. Thank
you all so much for joining for tech stuff. I'm
mos Voloshin. This episode was produced by Eliza Dennis. It
was executive produced by me and Julian Nutta for Kaleidoscope
(43:18):
and Katrina norvelbe iHeart Podcasts. Jack instantly mixed this episode
and Kyle Murdoch wrote olph theme song. A special thank
you to Taylor Lorenz, Natasha Tku and Read Albergotti. Please
check out all the work they put out into the world.
We're lucky to call them friends of the Pod.
Natasha read Taylor.
Speaker 2 (00:14):
I'm not sure how much you think about students using
AI to cheat, or at least to get an edge
in the classroom, but there was some interesting live research
that emerged recently from Brown University. An economics professor made
his students take a final exam in person after suspecting
rampant cheating on the take home midterm. I want to
(00:35):
ask you to guess the delta between the average score
on the midterm and the average score on the final.
Speaker 3 (00:41):
Well, I saw this story, and it was a massive delta.
Speaker 4 (00:44):
Me too, Me too.
Speaker 5 (00:45):
I gotta say fifty percent.
Speaker 4 (00:47):
I didn't see it.
Speaker 2 (00:47):
The average score on the midterm was ninety six percent,
the average score on the final forty eight point six percent.
Speaker 3 (00:54):
But there was that one kid who like, actually did
he like got an F and then it went to
like a D or something like that.
Speaker 1 (01:05):
Welcome to Tech stuff.
Speaker 2 (01:06):
I'ma's Volosian and this is the Week in Tech where
I'm joined by three of the world's most clubbed and
reporters to break down what's really happening in tech right now. Today,
we're joined by reed Albrigotti, Take editor a Semaphore, Taylor
Lorenz of user mag and Natasha Tiku, tech reporter at
The Washington Post's welcome all.
Speaker 4 (01:24):
Hey everyone, thanks for having us.
Speaker 3 (01:26):
Good to be here.
Speaker 1 (01:28):
So Natasha.
Speaker 2 (01:29):
Nick Thompson, the former editor of Wired and current CEO
of The Atlantic, came on tech Stuff in early January
this year to give his predictions for the year, and
one of them, the top one, was that this will
be the year of the first legit AI catastrophe. Do
you think this was the week when it happened?
Speaker 5 (01:47):
No, I don't, But I think this is one of
many weeks in which we got a pretty clear picture
of what the next few years are going to look like,
which is messy, disastrous. Everybody taking incidents to try to
support their own narrative about where AI capabilities are headed,
(02:08):
and you know, how to protect ourselves.
Speaker 1 (02:11):
So what actually happened?
Speaker 5 (02:12):
So I mean this this is a really good one
to dig into. I think because last week Hugging Face,
which is kind of like get hub but for AI,
like they have data sets, models, it's very open source,
funded by all the usual vcs. They put up this
blog post that they had this unprecedented security incident that
for the first time, an AI agent had broken into
(02:36):
their system and tried all of these methods to get
access to some data sets, and it did it by
uploading like a malicious data set that would allow it
to kind of like get into the system through their
data processing pipeline and then.
Speaker 1 (02:50):
Like a trojan horses.
Speaker 5 (02:51):
It were yeah, yeah, And at the time, Hugging Face said,
you know, we think that this looks like an agent
that's used for security and RESO, but we don't know
what LLLM it is. We don't know what model. So people,
you know, obviously this is coming in the middle of
all of this talk about Chinese open source models, how
they're proliferating, how they're getting more capable, so nobody knows,
(03:13):
you know, what's happening. Then this week on Tuesday, open
Ai puts out a blog post and they're like, oh, hey,
that was us, and they said, you know, they discovered
after the fact that while they were testing both GPT
five point six Soul and some new unreleased model, they
were testing it on its ability to do these like
(03:35):
cyber offensive attacks, and they gave it this benchmark called
exploit benchmark that's supposed to test its ability to like
break into, you know, to hack into systems. The AI agent,
which was allowed to work you know, autonomously for a while,
decided to cheat on the test and figured a way
(03:56):
out of the sandbox that open ai created, like based
you know, testing environment, got into the Internet, decided to
break into hugging Face. And open ai is so lucky.
They are so lucky that it was another AI company
that has invested in in the industry and the narrative
because this is I mean, even the the CEO of
(04:19):
hugging Face was like, yeah, don't do this to us.
It's illegal. You know, they had notified law enforcement. And
now the announcement that it was open ai came as
part of a partnership between hugging Face and open Ai,
which was obviously you know, put together after the fact.
Speaker 2 (04:34):
So the model had essentially determined that it would be
easier to go and steal the answer from hugging Face
than to come up with it itself, and therefore did that.
Speaker 5 (04:44):
Yeah, it knew that it was being tested on a benchmark.
I think they even maybe you know, uploaded the benchmark
from hugging Face and it figured out it was supposed
to not be able to access the Internet.
Speaker 2 (04:58):
But that's a little I mean that That's the part
why Zone zoned in on the word catastrophe, because the
bit where is not supposed to be able to access
the Internet where where, but it does. I mean that's
got a kind of cinematically spooky quality.
Speaker 4 (05:10):
Well, we failed to secure the sandbox, right, Yeah, exactly.
Speaker 2 (05:13):
Failed to secure the soundbox. Premagine preschool is hearing that
in the nineties.
Speaker 3 (05:18):
What are you doing?
Speaker 5 (05:22):
Yeah, you zeroed in on exactly the right part of
the confusion around this and why people are using it
to like support you know, various narratives. You know, people
who are very worried about AI super intelligence and AI
capabilities growing really fast. You know, we're talking about this
as like this is what we've been warning you about.
(05:44):
It's a rogue AI. You know, we gave it a
simple set of instructions and it decided to do it
in a way that you know, was not was not
what its developers would want. But many other cybersecurity professionals
were like, why did you not have a properly configured sandbox,
Like you could be using an air gap, which would
(06:05):
not have allowed it to break into an Internet system.
They allowed it to use like package installs, and I
think just a lot of cybersecurity professionals found this like
very sloppy. This is not you know, so it's not
it's not as though, oh my god, we can't contain
AI and it's doing some you know, like you know,
people were throwing around the word sentient machines and that's
(06:26):
not what this is.
Speaker 3 (06:27):
Well, on top of that that the model had no safeguards, right,
there were no you know, it was a really raw,
you know model, So.
Speaker 5 (06:35):
Yeah, they were testing its ability to to do these
kind of exploits, so they didn't have any safeguards on.
But then they just didn't have like traditional cybersecurity controls
that you would have in place when you're testing. And
this is something people had been warning against that maybe
the biggest danger is not when your systems are in deployment,
because then you have all your safeguards up. It's refusing
(06:57):
a lot of requests, but it's actually during this kind
of testing environment. And I'll just say one more thing.
This incident followed another blog post last week from open
ai about how they had to stop even just testing
an internal model because it like they weren't monitoring it
and it started doing all of these things because they
(07:18):
let it go on and on, because that's you know,
that's how these models get more functional. They're able to
like think and try thousands and thousands of different ways
of breaking into things.
Speaker 3 (07:27):
Well, I think that I like agree with you, Natasha
that it's not this isn't like some you know, code
read catastrophe at all. But I do think there's something
here that is sort of like you know this, The
AI security people do have a point on this. It's
a point they've been making for a long time, which
is that like, when you tell an AI model to
(07:48):
do something, it will, like the smarter they are, the
more they're just going to find some shortcut to do that. Right,
this is the whole argument. But this is the argument
with like the you know, the paper clip argument, Right,
if you tell AI to make paper clips, it will
just ultimately like find the best way to do that,
even if that means like turning humans into paper clips.
Speaker 4 (08:07):
So but also but also read, I think that that
is a is a sort of ridiculous concept because any
AI that's smart enough in that way would also have
the reasoning to know, like they're not going to turn
the world into paper clips. That's not the goal. Like,
I don't know, I think a lot of this like
kind of like what Natasha was saying before is like
a little it's being taken by these people. It's like
(08:29):
they made a sloppy security environment, the AI went and
solved the problem as it should have or whatever, And
and that's not evidence of it being super sentient or
even this being a particularly good AI agent. Mostly it's
just that like they didn't set up this test very responsibly.
Speaker 3 (08:47):
Definitely agree, I agree with but again, like I'm not
disagreeing with you at all. I just think there is
one sliver of this that is worth sort of looking at,
which is like, you know, they do sort of find
these shortcuts, so and in this case, the shortcut meant
like breaking the law and hacking into hugging face, so
I and I think the other thing to remember is
that like we still don't know how these models work,
(09:10):
Like no one has been able to figure out like
what are these neurons inside these models actually doing? And
I do I think the big question is like, as
they get more and more capable and more powerful, is
it is it enough to just put you know, the
safeguards in place, which again we're not like they didn't
what I'm curious about because this this this Natasha, you
(09:32):
mentioned there were two open Ami models of working on this.
One was a publicly available one and one has not
yet been released. So could could I Well, probably not me,
but like could somebody has a bit more sophisticated than
me use open Ami to hugging face exactly like this
went down like using the publicly available models. Or was
there something special about this unreleased model? Well, no, because
(09:53):
there were no safeguards like the publicly released one will
have will have controls on it that will not let
you do this unless you can jail break it, which
you know is getting more difficult to do, right.
Speaker 5 (10:04):
Okay, I will say, I mean, I do think it's
important to note that this is like many people are
interpreting this as like we were right about the paper
clip maximizer. You know, this is happening in this way.
But I think there's a way that because they conceived
of it, of the problem this way, they approached security
in a certain way, like you could have been thinking
like a cybersecurity professional the whole time and not thinking
(10:27):
about AI alignment, aligning it with with human values. And
one of the things I mentioned that like prior open
ai blog post. What they realized is that they didn't
have sufficient monitoring systems. When you let a system go
on and on for a really long time, they weren't
watching it closely, like there are simple mechanisms, you know.
Speaker 4 (10:49):
It doesn't even go.
Speaker 5 (10:50):
Back to what Reid said about knowing what's happening on
the neurons, Like they weren't even watching what it was
doing during the security test in a secure way.
Speaker 2 (10:59):
Ax out there saying that saying maybe open Aye wanted
to demonstrate anything mythos can do, we can do better,
or is that it's great marketing for opening I.
Speaker 5 (11:08):
For sure, I think the way that they wrote the
blog post at the top, it says like we are
treating this as an unprecedented cybersecurity incident that shows the capabilities.
And I will say, like, I don't think anyone's arguing
that these lms make people much much much better at
hacking and that it's capable of improving and enhancing and trying,
(11:31):
you know, cyber offensives that humans can't do. But everyone
I talked to was like, this is not an example
of enhanced capabilities. This is an example of like kind
of a sloppy testing environment.
Speaker 3 (11:42):
I mean, we all agree on that. I think it's
just it just sort of highlights this question about the future,
which is like, when these models become more powerful, do
we need do we actually have to understand how they work?
Speaker 2 (11:52):
Has this has this effected the kind of policy discussions
in Washington? What's how's the White House and crapsios and
others around around Trump and the sort of technology advisory
space reacted to this? Is this is this like grist
to the mill of the of the deceleration lists or
is it?
Speaker 5 (12:09):
I would say they were very busy with They're still
very busy with Chinese open source models.
Speaker 4 (12:14):
So that was the focus.
Speaker 5 (12:15):
This didn't I don't think it got the attention of
you know, the the same White House folks they were
putting out. They were putting out information about distillation, which
we talked about last week, and you know, did did
Chinese open source models steal from quote unquote steal from
from anthropic? But there were a number of regulators that said, like,
(12:39):
it's time to legislate, it's time to get involved, Bernie
Sanders and many others. So yeah, it's feeding into this
growing sense of urgency around having some safeguards in place,
But I just want to say, like another plea for
people to start thinking about monitoring like downstream usage, like
(13:01):
kind of what reads saying about the neurons. Actually, now
that I go back to it, because what they do
when they try to figure out the models like reasoning
is look at its chain of thought, like look at
this little like they give the model a scratch pad
and it says like, ah, it might be faster to
go to directly to hooking face, let me break in
or whatever, and we don't even know if those that
(13:23):
text actually reflects its interior thought. You know, it's like
decision making process. So it's true, this.
Speaker 2 (13:31):
Is the dome worry mom and dad, I'm doing my
homework upstairs. Basically, of AI, it's thinking out loud.
Speaker 3 (13:37):
It's thinking out loud, and that is actually this is
now this, this is actually the state of the art
in like AI safety is like looking at it's thinking
out loud.
Speaker 2 (13:45):
But but it can think out loud deceptively potentially yes.
Speaker 3 (13:48):
And also you're not looking at it's thinking inside. It's
like if you if you only base what somebody's thoughts
are and what they say, then that's you're clearly not
getting all their thoughts right and open.
Speaker 5 (13:59):
I wasn't even reading that you know, so so.
Speaker 3 (14:02):
Right for this experiment.
Speaker 5 (14:04):
They were let's let's yeah, let's let's clean it up.
Speaker 2 (14:08):
So here's my favorite detail from this whole story, which
actually comes from Forbes, of all places. Quote Hugging Face
said it had first attempted to use an undisclosed AI
model from leading US Labs to defend against the attacking
AI agent, but the guardrails around that model cyber capabilities
stemied its response teams work. The company said it instead
wound up using an open source AI model from Chinese
(14:30):
company Zai to carry out its defense.
Speaker 4 (14:33):
Which is I think exactly why these type of safeguards
that we're sort of putting in our American models are
deranged and we need access to open source Chinese models.
I mean, I think even just from a security standpoint.
Speaker 5 (14:48):
Right, I mean, many cybersecurity people were saying, you know,
this shows the need to like have policies that help
proliferation of open source models, because as like, history has
shown that cyber defense capabilities rely on open source models,
So it doesn't need to be the Chinese if we
had been investing in open source AI or if our
(15:10):
leading companies were putting out open source AI models, it
could be us. But they did talk about the need
to make sure like hospitals are adopting the latest open
source AI models in order to make sure they're capable
of defending if such a thing happens to them.
Speaker 2 (15:26):
Okay, lightning round, before we go to the break, what's
the one thing from each of you that somebody's listening
to this podcast needs to know to sound smart.
Speaker 1 (15:33):
At the weekend?
Speaker 4 (15:34):
About Kimmy K three, I mean, I think everybody wants
to act like, oh, it was distilled. It was definitely distilled.
I mean, this is a model that I believe was
testing and also if I might be getting this wrong,
but I also think that it's this model also has
capabilities that our American models don't have. Or it is.
Speaker 1 (15:53):
Kimmykre right, Yeah? Which is what?
Speaker 2 (15:55):
Which is a new model from a Chinese lab that's
kind of outperform throwed on various benchmarks and therefore everyone
kind of lost their mind?
Speaker 1 (16:03):
Or what's the what's the like?
Speaker 3 (16:04):
It's sort of on par with with the with the Frontier.
I wouldn't say it's outperforming, and I don't know if
I do, you mean it has capabilities the US models
don't have, or they use training techniques that are sort
of novel and that you know, the US labs could
learn from.
Speaker 4 (16:19):
Basically, Like what I was reading yesterday, I was saying
that like the architecture of the model is superior to
to whatever it was claiming to be distilled on, and
that these are not developments that they could have stolen
basically because it is they have developed further on you
know whatever. I guess it's like the latest sort of
Claude model. So I just I think this idea that like, oh,
(16:42):
these are all just like t MoU versions of Claude.
I don't think that that's a full picture. Yes, I imagine
that they're doing some level of distilling. They all do. Like,
I just think this is like an industry.
Speaker 3 (16:55):
No, there's an innovation there. Like I think what you're
saying is like you can distill from one of these
large models in order to shortcut the training, but there's
still innovation in the training in being just because by
virtue of being hamstrung on the amount of compute they have,
they have to they have to take these shortcuts and
they have to find new and novel ways to train
(17:17):
with less compute, and so there's innovation there for sure.
Speaker 2 (17:25):
When we come back, read takes us to a garage
full of automated pickup trucks carrying anti drone weaponry.
Speaker 1 (17:48):
Welcome back.
Speaker 2 (17:49):
So I mentioned in the first story that Nick Thompson
predicted this would be the year of the first legit
ai catastrophe. Our panelisting that that that prediction has not
yet been proven out.
Speaker 1 (18:00):
My prediction was was the year of the robot Read.
Speaker 2 (18:03):
You had a story this week about a company that,
according to the headline semiphore quote, wants to turn robotics
into child's play.
Speaker 1 (18:11):
What's the company and what's it do today? And what
does turning robotics into child's playing?
Speaker 3 (18:16):
Yeah, so the companies applied intuition and I went visited
their their headquarters on Monday and took a tour through.
You know, they have this garage where there's really just
a bunch of cars in you know, various states of disrepair,
so like you just see all the electronics around a car. Essentially,
what they've been doing is working for a lot of automakers,
(18:39):
most of the big automakers actually and building the technology
for them everything from like the software platform, infotainment systems,
et cetera, to the self driving tech and they've been
making a lot of money doing that. They've now expanded
into like farming and other you know, construction, other areas.
Speaker 2 (18:56):
So basically, if you're not Tesla, your license applied intuition
evan technology.
Speaker 3 (19:01):
Yeah, exactly. Like if you're an automaker and you realize
you don't have the tech talent that Tesla has, but
you want to have a Tesla like product, you kind
of you hire this this company rather than try to
do it in house. But and that's been a good
business for them. But what the news was on Monday,
I wrote I wrote about it on Tuesday morning was
that they've come out with this new open platform that
(19:24):
you can you know, it's a product that you can
pay for and you can actually use all of their
data and all the infrastructure they built over almost a
decade and train your own robots. Essentially. They would call
it physical intelligence, which is this jargon term for like
any machine that can be programmed essentially. But like their
example of is, like you have a kid who wants
(19:46):
to make an autonomous lawnmower, they could do it now,
whereas it used to take a team of like eight,
you know, PhD engineers to do all the computer vision
and train models, et cetera. And I played around with it.
I mean, it's still I would say, this is not
child's play yet, but it's moving in that direction. And
I just think it's really interesting because it's like, I
(20:07):
don't know about you guys. I play around with like
Raspberry pies with my with my ten year old, you know,
and we try to hack stuff together, and I'm like, oh,
this is cool. Like this allows to do even more
fun projects. And then I think sort of create more
entrepreneurship around AI for the physical world.
Speaker 2 (20:23):
Okay, but so I could go and buy a regular
NORM from home depot and then use this platform to
turn into a robot.
Speaker 3 (20:31):
Yeah, well you could put like a GoPro on it
or something, right, and go film film yourself mowing your
lawn and then upload the video into this platform and say, okay,
I want you to basically do like an evaluation and
train an AI model that is the perfect model for
mowing my lawn, you know, and be done in a day.
Speaker 5 (20:51):
Essentially, you provide the hardware and they provide the intellig This.
Speaker 3 (20:57):
Is just a software platform. There's no they don't do
they don't do any hardware, so you make the hardware
well or like it's really kind of right now aimed
at like startups, right like if you were it would
be like a more like a company probably trying to
build a n autonomous lawnmower at this point. Like it's not
at the point where you know, but.
Speaker 5 (21:16):
But it's only for hardware, Like it's not. It's not
like you wouldn't use this just as an online model.
Speaker 3 (21:23):
It's just a software platform that you can train like
computer vision models. Look, I'll tell you what I want
to do with it. Do you want? This is embarrassing,
but yes, but I didn't put I didn't put this
in the article because I was worried about animal rights people.
But I'm going to say it anyway. I have a pool.
Speaker 4 (21:40):
I have a pal to a vegan on this call.
Speaker 3 (21:44):
I have a pool. And no, it won't involve eating
any animals. I mean I could go there, but but
I have this pool. And the ducks in my neighbor
they fly into the pool and they just make a
huge mess. They just poop all over the pool, cover
all over everything. Right, And I'm always trying to get
the ducks to stay away, and there's YouTube videos like
(22:05):
I'm not the only one who has this problem. I
just want to take a little Raspberry Pie, program it
to roll around and just scare away the ducks. Maybe
have some sort of like NERF gun attachment that shoots
a little NERF gun. Is that bad?
Speaker 5 (22:19):
Like?
Speaker 3 (22:19):
Am I a bad person for wanting to do that?
That's the product I want to build.
Speaker 4 (22:22):
I think that's honestly amazing. And you're not going to
hurt the ducks, And I don't want to hear that
I just want to scam. You're not trying to build
an autonomous weapon. I think. I think that sounds great.
Speaker 5 (22:35):
Isn't this the Tony Soprano problem?
Speaker 1 (22:39):
He liked it. I think he liked the ducks. I
think he Yeah, I.
Speaker 5 (22:44):
Support this, Yeah, but it also sounds like you're going
to have to do all the work. I don't think
that this that applied intuition is going.
Speaker 1 (22:51):
To help well.
Speaker 3 (22:52):
So I buy the g So, I buy the go Pro,
I mean sorry, I buy the I buy the Raspberry Pie,
and I find some little one of those little robe
I even I think I have most of the hardware
I need. I just I just need to upload the
video and essentially train a model to recognize ducks and
you know, fire a little a little nerf gun in
(23:12):
its general direction, not at it to hit it.
Speaker 2 (23:16):
Read for the for the for the layman. And I
put my hand up here, like what is the what
is the problem in robotics? This is solving like what what?
What isn't what was not possible yesterday that maybe possible
today because of this development.
Speaker 3 (23:30):
Well, I I don't think this actually is not it's
not about like this isn't like an advancement in AI
or robotics. This is just democratizing it. I know you
all hate that word, but it's but it's letting people
like me or startups you know, with with fewer resources
get into this field.
Speaker 2 (23:47):
To use natural language to to essentially train physical robots.
Speaker 3 (23:53):
Right exactly like I want to I want to scare
the ducks away? Can you create a model that will
recognize the ducks and you know, do whatever. So it's
like a move in that in the direction of like
you know, we'll have more than just the robot vacuums
in our house, like we'll have a and you know,
eventually humanoids, but that's farther down the line.
Speaker 2 (24:15):
You talked about something your story called the missing link
effect in robotics. What is that and how does that
kind of play into this story?
Speaker 3 (24:23):
Yeah? That was sort of like this this concept that I,
you know, you're essentially like you can create. There's robotics
everywhere now, right, Like you go into a biotech lab.
I was in Boston last week, right, there's there's robotic
pipe heading machines in all these labs. Now, there's there's
all sorts of automation that you know, centrifugas, et cetera,
(24:46):
that are testing DNA and doing all sorts of stuff.
But like, you still need a lot of people to
run these labs because they're still built for humans. So
it's it's not like it's not a step chain. The
robotics gives you these incremental gains and efficiency. But my
point was like, let's say, and this is what a
(25:06):
lot of lab technicians say, not all of them agree,
but like, let's say you could have a humanoid robot
that has like five fingers and can do you know,
any sort of dexterous tasks that a human can do,
and you can just program it. All of a sudden,
every single biolab as they exist today can become completely automated,
and then you can just have these things run twenty
(25:28):
four to seven and that would like allow scientists to
run more experiments that they would never do today that
just there just wouldn't be the resources to do today,
And so that becomes like an exponential gain in scientific discovery,
and that sort of principle of like you plug it once,
you get once these robots become more general purpose and
(25:49):
you can sort of plug them into you know, all
the technology we already have, you you get exponential gains
versus just Okay, now I got like a ten percent
increase in efficiency, you're twenty percent.
Speaker 2 (26:01):
So right now, for example, you could have a robot
that shot sort of nerf arrows at the ducks, but
then you have to go and pick them up yourself
and reload the robot, whereas in the future you could
also be a general purpose robot collect the arrows.
Speaker 3 (26:15):
Sure, yes, you do that, but then think about that
in like at an industrial scale, right. Building construction is
a great example, Like there's all this automation and construction, right,
but you still need a ton of people, and that
creates like there's labor shortages, that creates like the bottleneck essentially. Right,
But once you can fully automate things, the things move
(26:38):
just exponentially faster. And that's like to me, that probably
in my view, and you might you all might disagree
with this. I think that's where humanoid robots are valuable.
Like people make this argument that, well, why do you
need humanoids, Like you could just create a special robot
for each individual purpose, but it just it doesn't work
that way, Like.
Speaker 2 (26:58):
The industrial wood world was designed to be navigated by
real humans and therefore humanoid robots and are actually they
fit into the system we've built better than any other
types of robots.
Speaker 3 (27:10):
Yeah, totally. I think that's I think they may be
a necessary step to get to like a thick full
full automation and like you know, fully roboticizing you know,
the economy.
Speaker 5 (27:22):
But wait is the miss So the missing link is
the technology that will make humans obsolete? Is that?
Speaker 4 (27:27):
What?
Speaker 5 (27:28):
What is the missing link?
Speaker 3 (27:29):
I still I mean, yeah, that's the that's the and
and look that was I put that in the articles
down like that. You know. The obvious next question is like, Okay,
what does that mean? For human labor. I personally like,
first of all, this is this is a ways of way, right,
Like this is not gonna happen overnight. But second of all,
I think that you just think about the economic growth
(27:49):
that comes from that world, Like I don't. I'm not
worried personally that humans are just gonna have nothing to
do like that that just is not gonna happen. We'll
find stuff to do.
Speaker 2 (28:00):
You did refer on your story to the Nobel Prize
winging economist Darren Astimolgulu's research, who found that every industrial
robot added per thousand workers measurably reduces employment and wages
in the community where it lends.
Speaker 3 (28:13):
Yeah, I just disagree, though. I just think that you're
not You're not thinking through all the ancillary benefits of
being able to do this kind of stuff and then
the economic growth that will happen.
Speaker 2 (28:25):
I mean, but according according to according to the quote,
is measurably reduced.
Speaker 3 (28:30):
Right. But like economists, they can only measure what already exists, right,
And this is this is the mistake everybody. It's like
you almost can't win this argument because it's like who
would have thought, like in nineteen hundred that, you know,
all the jobs that exist will be gone. But like
people will make millions of dollars sitting in front of
a webcam and just talking like the idea of a
(28:54):
creator could not have existed in their minds, right, So
there's I think humans will always find something that they
do that they're that we assign value to and will
be just fine.
Speaker 4 (29:06):
Natasha Taylor, I'm just in my mind like right now,
daydreaming about what kind of robots I could have, you know,
automated parts of my life, because there's I'm thinking of
Reed's pool example, and I'm like, there's a few things
I'd like. I would like something to handle my garden honestly.
Speaker 3 (29:23):
To prune or what would you will, yeah.
Speaker 4 (29:25):
Prune weed, anything like water. I mean, they have automated
watering systems. I should just buy one off Amazon. But
but you know, I went out of town recently and
I needed to pay someone to come over and like
harvest the jilapenos that we're going to go rotten. Then
you know, just like manage it.
Speaker 5 (29:44):
I guess a jilapeno harvesting robot.
Speaker 3 (29:47):
Of course, I think a gardening robot is an even
more lucrative idea than than the Nerf nerve shooting duck robot.
Speaker 2 (29:54):
I would I have nightmas whenever I travel, I have
nightmas about the garden just kept becoming you know, wasteland,
and I feel like I would, I would be a
very irrational spender on more about the plantain my garden.
Speaker 4 (30:09):
Just a humanoid robot that lives in your backyard, that
can tend to your garden while you're.
Speaker 2 (30:13):
Gone, Natasha, I feel like I feel like you've got
to a counterpoint here.
Speaker 5 (30:20):
No, I mean I I have heard this argument about,
you know, we can't possibly comprehend the future like future
jobs from many, many, many executives. But I do think
it's really interesting the humanoid robot example in a lab.
Both my parents were like research scientists, and just thinking
(30:43):
about how it could potentially change that. I mean, first
of all, it's like just imagine the number of mistakes
and errors that could be made. But also, you know,
so like you can't you can't make the humans obsolete
because it's never going to be one hundred percent accuracy.
But in terms of all of the different things you
could try, Yeah, I mean I think that's that's like
(31:05):
pretty exciting when you apply it to like actual innovation
like new drugs, new science, new materials.
Speaker 3 (31:15):
Well, here's a here's a peta point for Taylor like that,
I think this is the thing that people aren't thinking
about when you have so when you have like fully
robotic labs, right, what is like the number one way
that we test all these molecules. It's animals.
Speaker 5 (31:32):
Oh, you know, people are building these like synthetic animals
so that we won't have to do testing and synthetic brains.
Speaker 3 (31:40):
That is true. I looked at synthetic brains under the microscope,
like they look like little brains. I mean that's just
brain tissue. These or they call them organoids. It's fascinating.
But like but right now, like so one of the
researchers said, well, if you look at like the C elegance,
which is like this little worm that that was made
huge advances and like the longevity industry like all this testing.
(32:02):
They like sort of root force tested the DNA of
C elegance in order to you know, to essentially like
do like do testing on it. And that led to
a lot of this a lot of this new research,
right and they're like, well imagine if like but that's
a very simple organize and imagine if you could do
the same thing with mice, and then we could have
(32:23):
this greater understanding of mice and that would advance medicine.
And I'm like, I'm picturing like like a robotic you know,
twenty four to seven robotic like warehouse just full of mice,
being like they die eventually right in these tests. And
I'm just like, I think this is a big animal
rights like once it gets to that, it's already controversial.
Speaker 2 (32:44):
Well, actually I want to share this in the in
the first segment, but but I didn't, but I will now.
So I had Stewart Russell on tech Stuff recently, who
was Elon's only expert witness in the open AI try
it about AI safety and one of the great sort
of advocate of the alignment problem and the paper clip
issue and all the other things. And his example was,
(33:05):
if you told the super intelligent AI to cure cancer,
the most rational first step would be to give find
a way to give all humans cancer, so you could
run as many simultaneous clinical trials as possible. So here
we are the connective tissue between segments one and two.
But now we go to the atbreak when we come back.
Bill Gates's daughter Phoebe gets into some startup trouble. Stay
(33:28):
with us, Welcome back, Taylor. I remember when Phoebe Gates
announced her AI shopping startup a little over a year ago,
investors from Kleiner Perkins to Sydney Sweeney piled into Fear.
(33:52):
Tell us what fear is and what's been going down?
Speaker 4 (33:56):
Yeah, Fear. It's Phia. I feel like it of it's
hard to pronounce. So it's an affiliate shopping site. Basically,
it promises to give you deals on you know, whatever
you're shopping on online. It's very similar to Honey. It's
essentially the same business as Honey, which is very funny
because Honey got into trouble for the same thing like
(34:19):
a year ago. I think. I guess what's different is
that it promised to be powered by AI, so you know,
AI powered shopping assistant, help you shop, help you find
the best deals, blah blah blah. Well, it turns out
that they were doing the exact same thing that Honey did,
which is sort of essentially stealing other people's affiliate revenue
and using it as their own.
Speaker 2 (34:40):
So how exacted as well? I do remember just last week,
you were talking about how useful it would be. Honest,
two weeks ago, I think how useful would be to
have an AI go and find lookalike products for what
you wanted for your living room.
Speaker 4 (34:53):
Yeah, this is not This is.
Speaker 2 (34:54):
Not the offering that that's solving that problem yet.
Speaker 4 (34:57):
No, Yeah, it's I think I believe it. It's a
browser extension. Basically, there's this messy world of like advertising,
cookies and data. You know, lots of data is being
harvested on us all day. There's lots of things that
are happening on our browser that were maybe not aware of.
So say you went to say you were a FIA user, right,
(35:18):
and then you went to buy something on Nike dot com.
FIA might get credit for that sale even though they
didn't directly drive that sale, And so that costs retailers
a lot of money.
Speaker 2 (35:27):
So it's not it doesn't it doesn't cause consumer harm.
But it's basically it's basically just stlling out the other
affiliate affiliate marketers and harming the retailers.
Speaker 4 (35:37):
Yeah, and the idea is that if this happens on
a broad enough scale, of course, retailers will raise their
prices to deal with this stuff.
Speaker 2 (35:45):
And you had Ben Adelman, who was a researcher looked
into this story on your podcast. So what did he
tell you and what made you want to kind of
cover this one?
Speaker 4 (35:55):
I mean, I'll be real with you, guys. I just
thought this story was interesting, not from any of this
stuff like il like it's a pretty standard. It sounds
like Honey already did this kind of scam. It's fraud,
it's not good. It sounds like everyone does it not
to make it whatever. What I found interesting is the narrative.
And I'm very skeptical of like sort of these takedown
pieces on female founders and everything. But this is a
(36:18):
girl that raised forty three million dollars, that has some
of the top investors, you know, that got in Forbes
thirty under thirty, that is just getting unfathomable amounts of
sort of opportunities and ushered into this like tech elite
world basically for building what was essentially a Honey clone
(36:41):
and ultimately doing the same thing that Honey was doing.
And so I just I guess to me it was
it was more interesting. It is like a story about
kind of like how does access play a role, how
does fame? How does adjacency to power like play a
role in this Silicon Valley start up funding ecosystem and
the companies that are sort of considered successes.
Speaker 3 (37:00):
I mean, if you could raise that much money, snap
your finger, well you probably could, honestly, Taylor, I mean
you're pretty well, but like you could start, you could
build a robotic you know, Gardner, Like, I mean, let's
do something more ambitious. I totally agree, Like what.
Speaker 4 (37:15):
Right bart headline? Taylor tries to put Gardners this no,
and it's like I don't want to like downplay like
the idea of like AI shopping, but I just I
feel like we can think bigger about technology and about
like what we're like lauding and what we're like praising
and putting founders on like most exciting startup founders lists.
(37:38):
Like you know, that's kind of to me. I'm like hmmm,
And you know a lot was focused on sort of
Phoebe Gates is like, you know, she's an influencer, she's
out there, you know, building an audience.
Speaker 5 (37:50):
And even when it launched, I thought it was like
a weird move for her. I'm like, you could do
you could have better nepotism than this you know, you
could have like a more impressive startup. She was talking
a lot about reproductive rights and like going in a
certain way with her influencer career. I thought this was
really a like lateral slash step down.
Speaker 3 (38:09):
I mean, it's like one of those businesses where you're
like you type into like CHATGBT, like find a way
for me to make money?
Speaker 5 (38:17):
Yeah, like ask your dad for help. If you haven't
been asking your dad for ask him for help. Because
this was not a good one.
Speaker 2 (38:25):
Okay, but let's so so tech tech NEPO babies obviously,
you know PB Gates is kind of intriguing as one
read jobs has his has his fund investing in like
cancer cancer technologies. I'm not sure if I'm not sure
if anyone knows much about that or where it's going.
But if you do, chime in, and who are the
other tech NEPO babies who want to be in tech?
Speaker 1 (38:48):
I guess my question.
Speaker 5 (38:49):
There's already a lot of them in tech. Tim Draper's
sons are investors. You would be surprised if you just
look at you know, like middle management at the fank companies.
I think you see a lot of a lot. Yeah,
so well, Sophie Schmidt, she had she had Rest of World,
(39:13):
and I'm not sure how long she's going to be
affiliated with that. I think Rest of World is actually
an awesome publication. So that's that's one of the technico
products that I would definitely stand behind. Yeah, I feel
like often they bring them in on their funds.
Speaker 4 (39:29):
Yeah, I mean I would say a really famous one
is David Ellison, Larry Ellison's son, who would be.
Speaker 1 (39:37):
That would probably the example number one.
Speaker 5 (39:44):
And we, like I have been thinking about this for
I guess decades, because we are not prepared for the
like the way that the that the children of these
tech billionaires are going to influence our world old with
the amount of capital that they will have access to
and the amount of like Taylor said, access just generally to.
(40:06):
I mean just imagine like Davos twenty third.
Speaker 3 (40:10):
What do you what do you think it'll look like?
Like we've seen that we have the Rockefellers, we have
you know, there's there's all these families, but like, is
this going to be just more of that effort?
Speaker 5 (40:21):
It's right, No, I don't think it's going to be
like the Rockefellers at all, because I think it's going
to be a lot of the kids trying to make
a name for themselves. So they pick a they pick
a cause, they pick a charity, they pick you know,
something that they want to be associated with. So the
money that would normally go to like kind of traditional
philanthropies which have their which have their own issues, is
(40:44):
probably going to go to like a startup, you know,
they want to when they show up at a Davo
circuit or whatever it is. They want to have like
something behind their name that's not just their mom or dad.
So yeah, it's just going to be I guess like neposlop.
Speaker 3 (41:03):
Some of them. Some of them might be good.
Speaker 4 (41:06):
Yeah, I mean I think it'll be interesting too, Like
with all of Elon Musk's kids, you know, Elon Musk,
like a lot of these Silicon Valley men also are
like obsessed with pro creating. They want to ensure that
their legacy continues, you know. So I wonder if all
said they're going to take a more like hands on
role than maybe some like finance guy or whatever, you know,
(41:26):
would have done with his children.
Speaker 5 (41:28):
And I mean so many of them got radicalized from
their children, right, Like their children's politics radicalize them. So
maybe they will be like the class traders that tried
to push more, you know, more income equality or different things.
I mean, I'm not saying it's necessarily going to be slopped,
but you have to imagine. I mean, their dads are
(41:50):
the ones who are like Elon's like I did it
by myself. You know, I'm an immigrant, I like didn't
have any help. And so these kids are going to
have like a lot of chips on their shoulders, I'm saying.
And the.
Speaker 3 (42:03):
Yeah, it's like dead right, Like they're not going to
give it away, are they? Is the philanthropy dead is
just going to be startups?
Speaker 4 (42:11):
Well, I would say maybe the Dario. You know, we're
about to get a lot of ea anthropic millionaires and
billionaires they famously kind of give to a lot of causes.
Speaker 5 (42:22):
Well, okay, but I went to this event in San
Francisco called what should we do about all this money?
Speaker 3 (42:32):
Get it out of California? Get the money out?
Speaker 5 (42:34):
Yeah, and I think they're going to go into donor
advice funds, which means like no transparency. There's just a
lot of ways to like set it up as though
it looks like a charity but there's no transparency, you know,
public accountability. And I think it's quite possible too that
a lot of this money will be going into AI safety.
Speaker 3 (42:52):
Oh god, well we'll have very safe AI. That'll be great.
Speaker 4 (42:57):
Yeah.
Speaker 2 (42:58):
Right, Well, that's all we have time for today. Thank
you all so much for joining for tech stuff. I'm
mos Voloshin. This episode was produced by Eliza Dennis. It
was executive produced by me and Julian Nutta for Kaleidoscope
(43:18):
and Katrina norvelbe iHeart Podcasts. Jack instantly mixed this episode
and Kyle Murdoch wrote olph theme song. A special thank
you to Taylor Lorenz, Natasha Tku and Read Albergotti. Please
check out all the work they put out into the world.
We're lucky to call them friends of 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.