ML and AI with Sherol Chen: GCPPodcast 190

ML and AI with Sherol Chen: GCPPodcast 190

[MUSIC PLAYING] AJA: Hi, and welcome to episode
number 190 of the weekly Google Cloud Platform Podcast. I’m Aja, and I’m here
with my colleague, Jon. Hey, Jon. JON: How’s it going? AJA: Pretty good. So what do we have
in store this week? JON: So this week we get to
sit down with Sherol Chen to talk about ML and AI
and what it means to her, and really, really
amazing ML and AI trends. AJA: That sounds awesome. And if I remember from
the notes correctly, we’ve got a question
of the week about one of my personal passions, Ruby. But first, let’s spend some
time on cool thing of the week. What do you have for
us this week, Jon? JON: There are new AMD processes
coming to Google and Google Cloud, and there are a lot
of general purpose workloads that will see a pretty good
performance improvement. Around the same
price point, you’re going to have a higher
base frequency, which is really great, better compute
workloads for memory bandwidth, and they’re going to be roughly
around a 60% platform memory bandwidth increase,
which is kind of amazing. AJA: That sounds super cool. So one of my cool
things of the week is that Sara
Robinson on our team turned me on to this
Kaggle dataset that pulls data from Petfinder,
the adoption website. And I don’t know what
she’s doing with it, but I found the dataset and
I thought it was super cool, and it looks like there’s
a Kaggle competition around predicting
how long pets are going to stay in the shelter,
if I remember reading it correctly. But it has pictures
of adorable animals that were looking for homes at
some point in the last couple years, so that’s awesome,
because adorable animals are always awesome. So that is my cool thing
of the week number one. JON: Awesome. I’m pretty sure everybody’s
happy to see cute animals, and I know Sara’s been really,
really having a lot of fun with that dataset. AJA: That’s awesome. So do you have any more cool
things for us this week, Jon? JON: Yes. So one thing is now you can
stream data from cloud storage into BigQuery using
Cloud Functions, and this is really
amazing because you get really high availability
and auto scaling, and this creates almost real
time access to your data using Cloud storage, which is
really amazing because you get your information faster and it
responds more quickly to events because of Cloud Function. So it’s kind of cool. AJA: Yeah, Cloud Functions. I was a big skeptic on Functions
as a service for quite a while, but I have definitely come to
see the value and how useful it is, so that’s awesome. JON: And do you have anything
else for the cool things of the week? AJA: Yeah, actually,
I have one more thing, and this is, again, hitting
my passion project for Ruby. We now have App Engine
Standard for Ruby. I believe by the time this
episode goes out it will be GA, but it might still be beta. We’re still trying to sort
out when the GA date will be, but App Engine
Standard– we’ve had App Engine Flex for a while. App Engine Standard gives
us faster deploy times, and it gives us scale to zero
and a bunch of other stuff. I’m going to have a blog post
on my blog,, coming out later this
week, hopefully, maybe early next week, talking
about how to use it. But this is a huge, huge
thing for my Ruby friends that they can try
something on GCP that really works and
handles a lot of the hosting and stuff for you. I’ve tested it out
with an app I have that involves file uploads to
cloud storage, uses Cloud SQL, and it works really,
really well out of the box as long
as you understand a couple of the kind
of weirdnesses– we’ll go with weirdnesses–
about using GCP, and I’ll get to
one of those later in the question of the week. But I think it’s time to
go talk to Sherol now. JON: Sounds good. AJA: Let’s hear what
Sherol has to say about machine learning and AI. Hi, it’s Aja and I’m
here with Jon today, and our guest today is Sherol. Hey, Sherol. Can you introduce yourself? SHEROL CHEN: Hey, everybody. I’m Sherol. I work at Google
with all of you. JON: That’s awesome. So can you tell us
what you do at Google? SHEROL CHEN: Yeah. So I’m an advocate. I work in machine learning. Specifically I am working on an
awesome product called AutoML Tables, and what
it really does is it takes the concept
of machine learning and really breaks it down
to what data you have and what kind of predictions
you want to make. And these are just
really interesting tools that keep emerging, it seems
like, month after month. So I’ve been in machine learning
and AI for quite some time and served on a bunch of
different teams and roles in my time at Google. JON: That’s awesome, and
you’ve done some really, really interesting things, like I see
that you’ve taught a class. So you want to talk about
that a little bit and more about your background
a little bit? SHEROL CHEN: Yeah. So I would say that I
was born on a cold winter night or something like that. I think a really
defining moment for me was when I first started
playing video games, and I had this hero’s
journey experience that really got me through
the ups and downs of life and led me into
pursuing technology. It really stirred
a curiosity in me as to how do we
utilize these tools. And so my background
really took off, I think, in undergrad
where you were around all of these professors
and researchers, and I took a particular interest
in artificial intelligence, specifically because I felt
like games had improved so much in graphics,
but not so much in the logic behind the game. And then I went
on and did my PhD with leading expert in AI for
storytelling, Michael Matias, and from there I thought
I would be in academia for a good chunk of my life. And I did a lot of teaching
in the midst of going through grad school,
but then I also had an opportunity to work at
Google as a software engineer and saw kind of
a whole new world opened up and decided
to come to Google. And I ended up in the right
place at the right time. I was at Google Brain
when Sundar announced that we were going AI first. When I was at
Stanford teaching it was a time when a lot of
people were interested in AI, so I was putting together
the curriculum that was going to get used around the world. It just is something that I’ve
always felt an affinity for. AJA: Wow. That is super cool. Did I hear right? You partly got into AI because
you thought that NPCs and video games should be smarter? SHEROL CHEN: Yes, absolutely. If you looked at Final Fantasy
1 and then Final Fantasy XV, you’d see this huge
difference in the feel and the environment and
the sound and scene. But when you interact,
I just remember– I think for me, it
was Final Fantasy X– and you would look at
the maps of the world and it’s like a tube. You’re just going
down a pipe of point to point to point to point. On your way there, it’s
fantastical and amazing, but what about the actual
universe itself, you know? JON: Awesome. AJA: I think a lot
of our listeners probably know about AI and
have a good idea what it is, but a lot of people
define it differently, so what do you
consider AI, Sherol, so we can all be working from
the same set of definitions today? SHEROL CHEN: That is definitely
a question that comes up a lot. I’ve been interested
and studying and working AI for the last
10 years, so I’ve seen the shift in how
people think about AI. I would say that it really
stems from the ’50s and ’60s. That definition
is the one that I feel is historical
and has meaning, and there was that Dartmouth
summer school where the term AI was even coined. So I think that there’s
definitely historical meaning that I think is
valuable to consider. I also feel like
Alan Turing’s paper he wrote in the
early ’50s about can machines think was also posing
these really important thought experiments as to what do we
mean when we say intelligence. Now I think when I’ve
taught, whether it’s an 11-year-old or
a grad student, AI, the real question that you can
really relate anybody to is what does it mean to be human. Or a more specific
question is, what are the things that machines
do better than humans, and what are the things that
humans do better than machines? And that has always
been kind of a standard of how we develop and measure
artificial intelligence. So then when you think
about things that humans do, we can see, smell, hear, think. We have language. We’re creative. We’re visual. All of these things being
captured by technology creates these different
areas of research, whether we’re talking robotics
or natural language or machine learning. And machine learning has
really taken center stage in the last couple of years. I think it’s really
the move that made Sundar announce that
we were going AI first, because of all the advancements
that we were able to do. And that happened as a result
of all of the increased ability to compute, the amount of
data that we had, and just the algorithms working better. So AI has historical
meaningfulness. It also has different
depending on the context that you’re looking at. For me, my interest for AI was
believability and expressivity, so how we use AI to
extend human’s ability to communicate and express
themselves in story. And I think machine
learning and deep learning has been, in the last couple of
years, the big emphasis of what has shown to be more effective
than ever had been before. So if you’re talking
about AI, I think it’s worth reading the
history and understanding where it comes from. I think it’s really
important to not just think of AI as being
something really cool but really understand the
problems that AI is trying to solve, in which case you’ll
get a different angle of what AI is. And finally, if you’re
thinking about AI and what the buzz
is about today, it’s about deep
learning and machine learning because of the efficacy
of these new advancements in technology. AJA: So I think I know
the answer to this, but I’m not
confident, so can you tell me what deep learning is? SHEROL CHEN: Yes. So machine learning
is something– Alan Turing, back in the 1950s,
he said, machine learning– can machines think? Can machines actually acquire
knowledge on their own? That’s some of the
earliest published concepts around machine learning. I think one of the earliest
programs that was actually made was actually, a lot of
these early AI programmers were psychologists
or they were in these interdisciplinary roles
of technology and psychology. And I believe that the first
machine learning program was actually a checkers program
where somebody was wondering, how could I build
checkers program that could beat me at checkers? And so it would apply
some sort of reinforcement learning to be able to figure
out how to win at checkers. That actually happened
not that much– it must have been a decade
later or something for which that happened. So it wasn’t something
that was in our recent last five to 10 years. Now the problem is
that there’s been a lot of randomness and
unpredictability of having the machine just kind of
process information on its own without the guidance
and the formalities that we lay out, like the
rule-based systems that we typically use. I think, hand-in-hand,
there’s an intuition of, oh, we can kind of guide the
logic of how the machine is going to break down. What’s the best decision
given a particular situation? There’s all these
different states, and this clearly is
the more ideal state and feed all the
information for that. So that’s machine learning. And again, not everybody agrees
this is the way it works, but there’s AI, and then within
AI, a part of intelligence is the ability to learn, for
which machine learning is just the computer acquiring its
own understanding of some sort of process or decision. So with deep learning
and machine learning, there is this concept
of neural networks. And neural networks
are not the only way– this idea of a
computational neuron is not the only way
machines can learn, right? There’s other ways. Decision tree
learning, for example. But with neural networks,
you’re basically balancing a bunch of weights. You’re given
information, and it goes through these mathematical
calculations in neurons that balance a bunch
of weights to create this model, or this
mathematical representation, of how, given input
A, you get output B. How it gets deep is
that it’s very limited if you stick to these
purely [INAUDIBLE] or linear operations. Deep learning really introduces
the ability for the computer to remove and hide some
of that information and pick and choose
what it’s going to do, and it introduces an even
bigger layer of obscurity, but it gives the
machine more power. What it does come at the
cost of is understandability of what’s actually happening
when it makes that decision. There’s AI, there’s
machine learning, and then there’s deep
learning, and then within that is neural networks, and then the
hidden layers are the ones that make it a deep learning system. JON: Awesome. You said something
really interesting that I want to ask about. You mentioned that, early on,
AI was pretty much psychiatrists and someone in computer science,
like the intermediary track. But I’m curious, for
AI and ML, where’s that line drawn between
what decisions a human makes and the type of models that
you have to actually build and what the machine
actually does? SHEROL CHEN: Yeah. I think that humans
are imperfect, so when I asked the question
in a classroom, what are things that humans
do better than machines? What are things that machines
do better than humans? And then a common
debate is driving. So who’s going to be a
better driver, or what is going to be a better driver? Is it going to be the human or
is it going to be the computer? And you’re kind of torn. There are a lot of
things that humans can do that we take for
granted, but there are also just clear weaknesses
that we have, right, like when we’re
tired, when we’re inebriated. There’s different
things that prevent us from functioning
at full capacity. And so a lot of our traffic
laws are built around this idea that you’d have all your correct
sensory functions in order. When we think about how we
create policies for traffic laws, for example,
they exist that you need to be able to
see, for example, to be able to drive a car. But there is some
margin of, well, you don’t have to have exact,
perfect, immaculate 20/20 vision. You can have a little
bit less than that, let’s say, and then there’s
some amount less than that. And all of that needs to be
researched and considered to be able to protect the
safety of society as a whole. But I don’t think that’s
an easy question to answer, because when I talk
about self-driving cars, if we talk about it in Silicon
Valley or the Bay Area, we’re thinking like highway
280 where it’s wide and open and the lines are very clear. But I’ve taught in
India before, and when I show them the video of a
self-driving car in California, yeah, that does
seem like terrain that a machine can navigate. If you show them the streets
of some of the places in India, I just don’t even know
what kind of input– you’d just have to have a
very fast, reactionary state of processing. It’s a complicated
system, but I don’t think that it’s anything new. I don’t think that how
societies function up to now– we’ve always had new technology. How did we introduce the car? How did we introduce
the printing press? All of these things
have required us to adjust the way we
understand and see the world. JON: That’s super insightful. I have another question. You mentioned
self-driving cars and we started talking about
robotics, but what about the rest of technologies,
such as education and medicine, potentially. What impact would AI
and ML have on that? SHEROL CHEN: Yeah. So I think that that word,
impact, is very appropriate. I was recently asked to be
a judge for the Google’s AI Impact Challenge where a lot of
grant money was awarded to 20 recipients and they all came
up with proposals on what kind of impact AI could have–
social good impact– in the world. And one of the things
you brought up, medical. A lot of these grants were
awarded to medical use cases. Being able to pass
through medical records, which has also evolved. We have all this data. How do we use it? How do we use it to do
preventative things? How do we use it to help,
not just humans, but animals, like agriculture technology. A lot of people in
areas where they’re focusing on food
and plant livelihood are able to now use technology
to be able to figure out how to best cultivate resources
for maximizing food and health and safety and things like that. And the other thing is that,
now that we have Android phones, we have the ability to do a
lot of things with the camera and we have on device. We could build models
and train them on device. We can take this relatively
inexpensive medical device through the phone and
go to areas in the world where you can diagnose
infections and be able to prescribe the right
antibiotics or medicines. So these are things
that you would be able to multiply the
abilities of these high in demand skill
sets, like medical, but also other things like
the environment or agriculture or social services, like
education, and the ability to have coverage in areas
that we didn’t have before. JON: So you mentioned
the AI Impact Challenge. I happened to see– I’m not sure which year this
challenge has been going on, but they announced the
winners at Google I/O, which was really awesome. I think you mentioned that you
had something to do with that. You were a judge. But I’m curious, when you
think about Impact and AI, what about the ethical
parts of AI and ML? Because I would
imagine that some of these contestants or
participants in the challenge would have to think
about these things when they’re actually building
out any of their solutions to real world problems. So I’m just curious,
what do you have to say about using
AI and ML ethically? SHEROL CHEN: That was definitely
one of the considerations when reviewing. And often you get asked, whether
you’re speaking at a conference or at an event, that people
will ask, how can I use AI? We work with enterprise
customers a lot, and they’re using AI to
help better their business or to help make better choices
in directions that they’d like to take. So definitely AI
is a powerful tool to be used in these processes. When I was looking at these
applicants, a lot of this stuff came to mind. I think, especially when
you’re dealing with something medically related,
yeah, it would be great to get an
awesome medical dataset, but how do we protect
the people that are being represented in this data? And then say you’re using
Google Assistant to check out how someone’s mental
state of being is. How do you make sure that
that is not infringing on that person’s privacy? So these are things
that really do need to be not
reactively thought about, but proactively thought about. And so that’s something that– I think it’s arguable that
we’re kind of keeping up with the move of
technology, but I also feel like there’s
always more that can be done in those regards. AJA: So what kind of tools
are available to people who want to start
learning about AI or start trying to train
models, and what kinds of tools do you work on at Google? SHEROL CHEN: OK. Of course there’s buzz around
AI, and people are like, it’s just kind of the hype
of the moment or something. But I think it’s true that
this technology is going to continue to
improve, and it’s going to be a big part of society. And I think similar, maybe
not in the exact same way, as how the printing press
changed baseline skill sets that people needed. I think if you look at reading
literacy 100 years ago, it was probably like 90%
of the world couldn’t read, and now it’s flipped. Now 90% of the world can read. So I think that if
we can stay on top of what’s going on in
the world in technology, I think that’s very important. So for me personally, I try
to read about what’s going on, what other businesses are doing,
what other research groups are working on. You try to keep track
of the big conferences like [INAUDIBLE]
and ICLR and ICML. I tweet a lot, just a lot of the
current, interesting highlights of the moment in
machine learning and AI. But outside of just
keeping up as to what is being presented through
conferences and media, I think a really good resource
for learning the basics of it and just understanding it
from a technical standpoint is the machine
learning crash course. I’ve gone through it twice. One time I went through
it in a guided classroom, and then I just went
through it by myself and just looked at
every unit and worked through a lot of the
basics and fundamentals through that crash course. And it uses
TensorFlow and Python, so you have to have
a baseline in that. But I thought it was one of
the best, most up to date representation of what AI is
being most used right now. Now, there’s other things. If you’re more interested in
exploring open source projects, there’s a variety of open source
machine learning projects. TensorFlow is open source. Project Magenta, which is
the team I was on at Brain, has a lot of creativity
based, so music, art, generation-based models. And so you could provide
your own datasets. You can try to build your own
front end for these tools. And I think that would be
in general type of tools that you would use. Now if you’re a
developer and you’re just like, how can I
use machine learning, I think it’s a matter of what
you’re trying to do with it, but there’s a whole range of
types of tools, let’s say. So you could have
pretrained models, so as you would get from talking
to Google Assistant or Siri or Alexa, a lot of
this speech to text is ongoingly being
worked on, and so that would be use
case that you wouldn’t need to build from scratch. You could use Google’s
pretrained speech to text API, for example, to be
able to parse that. Or things like natural
language, or things that– general image recognition
type use cases, like maybe sentiments,
whether someone’s happy or sad in a photo. Those are things that
are general enough that can be continually worked
on by these bigger companies. And if you want
something more custom– so if, say, you
have your own data, you have a list of
customers or you have a bunch of
retail data, you want to be able to train
your own custom model, that’s where you use auto
machine learning, which still lets you customize a lot
more than you would with some of the pretrained APIs,
but at the same time you’re not really doing much of
the machine learning algorithms yourself. Finally would be just,
you could also always code it up from scratch. So I think that there’s a
whole range of different tools that you can use to get a
better understanding of machine learning, and I do
think that, as we’ve seen, as we’ve discussed
and as we’ve been explaining with the AI Impact
Challenge, I think having the right problem
also really matters if you’re getting serious into using
machine learning for real world use cases. JON: So Sherol, we like to
do this thing on our podcast where we ask our guests,
what is something cool or awesome you’ve
seen done with AI or ML that you’ve seen done recently. So you want to
share that with us? SHEROL CHEN: So yeah, that’s
a really good question. I would definitely say
one of the biggest things to pay attention to is AutoML. So for auto machine
learning, it’s really a step in
between having the API or having the model already
pretrained and ready to go. You just give it inputs
and it gives you outputs. It’s kind of the middle ground
between that and building everything from scratch. And so what you do with
auto machine learning is the ability to be able
to take data that you have and create a custom
prediction model that suits your needs specifically. Now, I definitely
think, as we talked about earlier with ethics,
that it is, in a sense, democratization,
but I don’t think it replaces really understanding
how machine learning works. So while I think
it’s a great tool and it’s going to give a
lot of developers access to be able to use machine
learning in their applications or tools that they
want to build, I do think it’s still
important to understand how you would do it by
yourself if you needed to. Because as much as
we want it to be easy to use, at the end
of the day, some of it is still being maintained
by a third party. So I do think it’s magical. I think it’s great. I think that there’s a lot
of science behind that magic, and it’s great to
cultivate a curiosity to understand how it all works. But yeah, I would say that
the coolest thing right now is this ability to just see
something like AutoML tables. I worked on the
promo video for it and I showed some of
my developer friends after it was out, and they
were just like, on one hand, it’s like, yeah, of course
this should be a thing. Of course you should
be able to just select a column in your
spreadsheet and be able to predict the values
based off of the other columns. On the other hand,
it’s like, wow, I can’t believe this
is really happening. We’re actually at this point
where we can see this stuff. So yeah, I think that that– I mean, there’s so many other
things going on right now. I think the AI
Impact Challenge, I would check out the 20
projects that got funded just to see what the latest– and these are vetted by Jeff
Dean and a huge team of experts as well, so there’s so much. JON: So thanks, Sherol,
for joining us today. And we’re running
out of time, but I wanted to know if
there’s anything that we missed or do you want
to let our listeners know where you’re going to be in
the upcoming months? SHEROL CHEN: So yeah, it
was a great, great pleasure talking to you both. I love talking about
artificial intelligence. It’s a great passion
of mine, and I think it’s great
to hear and discuss technology and the potential
for where it’s headed. If you want to
follow me on Twitter, you can find me @ffpaladin. FFpaladin is my Twitter handle. I also blog occasionally
on, so you can find me on that blog. If you want to check out that
promo video for AutoML tables, that was a project that
I worked on as well. So other than that, everything’s
basically on Twitter these days, so I
would just use that. AJA: Yeah, that is def– JON: Awesome. And is your Twitter handle,
ffpaladin, for Final Fantasy Paladin, by any chance? SHEROL CHEN: Yep. I’m going to date myself,
but I was from the AIM days. That was my AIM name,
and I just kept it. It became my Twitter handle,
and it’s been me ever since. JON: All right. Well, that’s going to
do it for us today. Thank you for joining
us again, Sherol, and we’ll see you
on the next episode. Thanks, Aja. AJA: Awesome talking to Sherol. JON: Sherol has a really
big passion for ML and AI, which is something that
we really, really love to see at Google. AJA: Yeah, it’s
always great when we get to talk to folks about AI. It makes me happy. JON: Let’s talk about our
question of the week, Aja. You mentioned that you can run
App Engine Standard and Rails, so how do you handle migrations? AJA: So this is one of
those things that comes up every time we deal with
Ruby because Ruby has a web framework. There’s multiple,
but a lot of folks use Rails or use Rails like
Frameworks based on rack. And we had to handle this when
we originally did App Engine Flex, but you also run into
similar stuff with Django, and someone on our
team has worked out how to solve that issue. But the answer to how
you run migrations is with a gem we have
called the App Engine Gem. There’ll be a link to
it in the show notes. You install it
like any other gem, and it basically gives you
the ability to run Rake tasks. Rake, for those of you who
aren’t familiar with the Ruby ecosystem, is like Make, but for
Ruby, so it starts with an R. And this allows you to run your
arbitrary Rake commands, things like Rake D Migrate,
Rake D Create, against your
production environment. And it does it
using Cloud Build, and it’s really
awesome because it lets you use most of the same
workflow that you’re used to, but it does things in
a secure and safe way. And it’s fantastic and
it’s pretty seamless. I’ve had really, really
good luck with it when I’ve been using App
Engine for applications and production. So that exists, and that
answers the question. Probably the most
common question we get about Rails + GCP is,
how do I handle migrations? And the answer is
the App Engine gem. And I want to thank
my colleague Daniel Zuma for doing most of the
development work on that. We wouldn’t have it without
his passion for Ruby and Rails. So that is our
question of the week, and my somewhat wandering
answer about it. Now for my favorite part. Where is everyone going to be? Jon, I hear you have
some stuff coming up? Want to tell us about that? JON: Yeah, sure. So I’ll be flying to Seattle. Maybe I’ll get to
stop by in the office and hang out with you for a bit,
but I’ll be in Seattle for PAX and it should be really fun– PAX Dev and PAX West. I’ll be talking at PAX
Dev with Mark Mandel, and I’m really
excited because we’ve been working pretty hard
on the demo for that talk and it’s going to be really fun. AJA: Yay. JON: Fun being the keyword
considering that this is going to be a game demo. AJA: Games. JON: I’ll also be at
the internal Google Game Summit, which is
coming up in September. I’m pretty excited
about it because we get to hang out with
a bunch of people who are really
passionate about games and seeing if we can really make
this games thing really happen at Google. After that, I will be
traveling to Canada to hang out with my twin
brother and celebrate his marriage before it happens. AJA: This sounds interesting
and strategic and awesome. Are you going to have poutine? Get yourself some
massively awesome fries while you’re in Canada? JON: I’ve never had
it, but a lot of people were telling me about
the food in Canada, so I’m going to check everything
out as much as possible, because we have
quite the itinerary. AJA: Oh, I’m sure you do. JON: And Aja, where
are you going to be? AJA: I’m boring. I’m actually not boring. I just don’t have a lot of
travel planned right now. I’m focusing on spending
some time at home and getting stuff
sorted at work. So the best place
to catch me, if you want to see what I’m up to, is
catching me on the internet. If you happen to be
local here in Seattle, I’m usually at Seattle
RB every Tuesday, which is our local Ruby
brigade meetup, and I will probably be popping
into various other meetups over the next couple
of weeks, and I will be sure to tweet about that. And if you to catch
me on the internets, I am the_thagomizer on
Twitter, and on the internets is my blog. So yep, that’s where
I’m going to be. JON: Sounds fun. Well, I think that’s going to
wrap it up for this episode, and we would like to thank
you all for listening, and we’d like to
see you next week. AJA: See you, Jon. JON: See you, Aja. SPEAKER 1: A pronoun
that is usually defined is used to indicate a person,
thing, idea [INAUDIBLE] SPEAKER 2: Oh my gosh. I don’t know how
to shut that off. AI is interfering with
our interview right now. AJA: Yeah. The meta level of having
AI interfere affair with the AI interview. JON: That’s a great tidbit
right at the end of our episode. AJA: It’s apparently giving
a grammar lesson today. SPEAKER 2: I don’t
even know what it’s– SPEAKER 1: The adverb that is
usually defined as [INAUDIBLE] SPEAKER 2: [INAUDIBLE]
has been disabled.

Posts created 41002

Leave a Reply

Your email address will not be published. Required fields are marked *

Related Posts

Begin typing your search term above and press enter to search. Press ESC to cancel.

Back To Top