Latest / The Learning Corner by Precursor / Episode #86: Paying For AI Twice, Sierra's Bet On One AI Agent, Data's Next Trillion Dollar Category, NY Halts Data Centers

Transcript

Episode #86: Paying For AI Twice, Sierra's Bet On One AI Agent, Data's Next Trillion Dollar Category, NY Halts Data Centers · The Learning Corner by Precursor

• 0:00
  We talked about this before.
• 0:01
  This is the first tech wave in my career where I've seen active organized opposition.
• 0:05
  Other things people have been like, I don't know.
• 0:07
  I'm neutral on mobile or mobile's good.
• 0:08
  This is people are trying to stop something from happening as opposed to undoing something
• 0:13
  that's already happened.
• 0:14
  I think tech needs to take these things very seriously and ask ourselves a question, what is
• 0:18
  in it for the average person?
• 0:20
  It's the week of July 13.
• 0:22
  Welcome to the Learning Corner, a Precursor Ventures podcast.
• 0:28
  Welcome to the Learning Corner by Precursor.
• 0:30
  I'm Mia Farnham, and I'm joined by our GP and managing partner, Charles Hudson, where we will
• 0:34
  be talking through the articles and Goodreads that caught our team's attention over the past
• 0:38
  week.
• 0:38
  Our first Goodread this week is a post on x from Satya Nadella called the reverse
• 0:43
  information paradox.
• 0:45
  Satya builds off of economist Kenneth Arrow's classic paradox where a seller of information
• 0:50
  risks giving it away for free just by trying to sell it.
• 0:53
  He argues AI flips that script.
• 0:55
  So now it's the buyer who risks giving away their most valuable knowledge, the contacts,
• 1:00
  the corrections, the internal know how just by using the tool that they paid for.
• 1:05
  Every time an employee corrects a model or feeds a proprietary information, the knowledge
• 1:09
  quietly becomes part of what the AI vendor learns.
• 1:11
  And over time, the vendor ends up knowing more about your business than you know about them.
• 1:16
  His fix is that every company needs a hard boundary around its own data, feedback loops,
• 1:21
  and institutional knowledge, similar to how a patent lets an inventor disclose an idea
• 1:25
  without simply giving it away.
• 1:27
  He points to Alex Karp's line about customers wanting to own their compute models and data
• 1:31
  stack as proof that this is already on enterprise buyers' minds.
• 1:35
  I wanted your thoughts on this more generally, but we're obviously thinking about this
• 1:38
  Goodread from the standpoint of b to b companies wanting to protect their know how
• 1:43
  while still capturing efficiencies from AI.
• 1:45
  But if you flip it to the standpoint of an employee sitting inside one of these
• 1:49
  organizations, if using these tools is making me faster and better at my job, does it
• 1:54
  actually matter what the seller is doing with my data on the back end?
• 1:57
  Is this a trade off worth carrying about, or is it just noise compared to the productivity
• 2:02
  gain?
• 2:03
  And when is that start and stop?
• 2:04
  I think the theme of this week's learning corner is, like, be careful what you train and
• 2:08
  how you train it, because there's some other things we'll talk about later that are related
• 2:12
  to data labeling and training.
• 2:13
  First of all, the fun thing, if you're Microsoft, it's like, well, we win anyway.
• 2:17
  We're selling a lot of compute capacity to some of the big hyperscalers and hosted closed
• 2:22
  source models.
• 2:23
  And if our customers decide they wanna use open source models, we're happy to sell them compute
• 2:27
  and hosting for those open source models too if they're not gonna run them truly, truly
• 2:31
  locally.
• 2:31
  So then if you're Satya and Adele, you're like, as long as people are using AI and they're
• 2:35
  using our infrastructure, I don't really care whether it's OpenAI, Anthropic, Quen, Kimi,
• 2:40
  whatever model they wanna use.
• 2:41
  This to me feels like it was the inevitable pivot back from let's just trust the closed
• 2:46
  source hosted models.
• 2:48
  Yes.
• 2:49
  We have these existential concerns about security, but what's our alternative?
• 2:52
  Your alternatives in the beginning were use the hosted models and hope that they weren't using
• 2:57
  your data to train the system that you checked that little box.
• 2:59
  And now the open source models are getting better, and people are getting more comfortable
• 3:03
  saying, hey.
• 3:03
  Maybe the right thing to do we talked about this with Kirkland and Alice.
• 3:06
  Maybe the right thing for us to do is to make the investment in training our own models.
• 3:10
  It's funny.
• 3:10
  I've been doing this this weekend with a fun side project, a video related side project,
• 3:15
  training a video model, a version of Quen on a bunch of video assets that I have that I'm
• 3:19
  trying to analyze.
• 3:19
  And it's fun, and it's more for cost for me than for security.
• 3:23
  And I think we're gonna end up in a world where people are gonna use it's the old you don't
• 3:27
  drive a Ferrari to the grocery store analogy.
• 3:29
  Like, people are gonna end up using the frontier models to do the frontier tier work,
• 3:32
  work, the really, really hard stuff where you want the most sophisticated models.
• 3:36
  There'll be some things for privacy, propriety, whatever people will want to keep inside their
• 3:41
  own firewall, whether it's actually worth protecting or not.
• 3:44
  It's probably more about a feeling of safety and a feeling that we're not this information
• 3:49
  than it is that, like, having it out there in the model would really make.
• 3:51
  A difference is whether you're gonna see a pivot back towards maybe we should control and
• 3:55
  manage AI.
• 3:56
  It's funny.
• 3:56
  I told him our portfolio company founders this.
• 3:58
  AI is just becoming part of corporate infrastructure again.
• 4:01
  And for a while, it was this, like, Wild Wild West rogue thing where people are like, well,
• 4:05
  we don't need controls or compliance or training or guardrails just like spend, spend,
• 4:09
  spend, and test, test, test.
• 4:11
  And now I think people are like, well, what if we just treated this like other parts of
• 4:14
  infrastructure where we care about ROI, where we care about compliance, where we care about
• 4:18
  data security, and where we weigh the benefits of, like, openness versus this.
• 4:22
  And I think the the pendulum is gonna swing back towards treating AI more like corporate
• 4:26
  IT.
• 4:27
  It will not be fun for the people who are, like, out here doing all these crazy cowboy
• 4:30
  things.
• 4:31
  Yeah.
• 4:31
  Well, that was my next question.
• 4:33
  Whose job will that be to determine what's absolutely has to be secure and compliant, and
• 4:38
  then what can you explore with frontier models and just kind of run amok on your own?
• 4:42
  I'm just sure in two years, you're gonna hear about an engineer who wasn't using all their
• 4:47
  tokens who, like, swapped their tokens with some other b l's of their team who needed them
• 4:51
  in exchange for a vacate.
• 4:53
  It's gonna get worse.
• 4:53
  Ordering.
• 4:54
  Yep.
• 4:54
  But I think the CFO is gonna have a big role in this.
• 4:57
  And you're already seeing what I would describe as brute force attempts at this.
• 5:00
  That company is just saying engineers can only spend x amount in tokens per month.
• 5:04
  I'm like, but what if one person's really, really efficient at getting value out of
• 5:08
  tokens?
• 5:08
  Maybe that person should be allowed to spend more.
• 5:10
  And the person who's just wasting them translating PDFs into PowerPoint shouldn't get
• 5:14
  any tokens.
• 5:15
  Yeah.
• 5:15
  But we're not we're not at that level of sophistication and nuance yet.
• 5:18
  No.
• 5:18
  We're still token maxing.
• 5:19
  Yeah.
• 5:20
  One last piece I wanted to touch on this.
• 5:21
  Another comment that was in the threads a little bit further down the chain was how are
• 5:26
  workers' IP protected from the enterprises.
• 5:28
  So Yeah.
• 5:29
  One click below that.
• 5:30
  We talked about this a little bit ago when we were reading the takes on if you got to take
• 5:34
  agents from one job to the next with you, if you built them at a job.
• 5:37
  But the knowledge that you bring to a workspace and the know how, that's training your
• 5:41
  enterprises model on.
• 5:43
  How do you as the employee get to ring fence?
• 5:45
  What's trained on that versus what you get to keep?
• 5:47
  Everybody involved in training is telling the people doing the training, don't worry.
• 5:51
  You'll be fine.
• 5:51
  You know, all the data labeling people are like, don't worry.
• 5:54
  You labeling all this data by your profession, you'll be fine.
• 5:57
  The companies are telling their employees, you teaching the AI how you work, don't worry.
• 6:00
  You'll be fine.
• 6:01
  I don't know if everybody's telling the truth.
• 6:03
  More to come.
• 6:04
  Well, our next good read is a blog post from Neil Rahili at Sierra called AI Pilling Our
• 6:08
  Company, Lessons Learned.
• 6:10
  After their engineering team started seeing five x productivity gains from running coding
• 6:13
  agents in parallel, Sierra set up a six person AI acceleration team to figure out what it
• 6:18
  would actually take to get that same boost across the entire company, not just engineers.
• 6:23
  They started by building a separate agent for each department, one for support, one for
• 6:27
  sales, one for data, but found that setup actually got in the way since the most valuable
• 6:32
  work happens across teams, not within one function.
• 6:35
  So they collapsed everything into a single agent, called it pine cone, one that persists
• 6:39
  across an entire project instead of resetting after each request since most real work doesn't
• 6:44
  actually get done in one sitting.
• 6:45
  It unfolds over days or weeks.
• 6:47
  One of Neil's bigger points is that the agent should really be thought of as just the
• 6:51
  interface.
• 6:52
  The actual system of record, your CRM, your ticketing tool, your code base still lives
• 6:56
  underneath it, and an agent is only as good as the quality of the system that it's pulling
• 7:01
  from.
• 7:01
  Since launching, Pinecone has run over 75,000 sessions and now opens 70% of their engineering
• 7:08
  pull requests.
• 7:09
  But Neil is careful to point out that usage numbers alone don't prove the tool is actually
• 7:13
  making outcomes better, just that people are using it.
• 7:16
  So Sierra admits they can't quite measure whether Pinecone is actually improving outcomes
• 7:19
  versus just generating more activity.
• 7:21
  That's coming, like, from an agentic company themselves
• 7:24
  Yeah.
• 7:24
  That they build for a business.
• 7:25
  Would you have expected more from that experiment from them, or do you think the
• 7:29
  insights were kind spot on?
• 7:30
  I appreciate the honesty and the willingness to not AI wash what actually happened.
• 7:36
  I really enjoyed reading the article, and I think what they're doing is this everyone says,
• 7:40
  oh, there's no ROI.
• 7:41
  I'm like, yeah.
• 7:42
  Because the gains aren't broadly distributed enough across the company to show up.
• 7:46
  And I think in the same way we've talked about companies that are beginning to take these open
• 7:51
  source models and use them instead of the closed source models, I think you're gonna see
• 7:54
  more companies that say, you know, we have two choices.
• 7:56
  We could hire a very expensive consulting firm to come in and do an AI transformation, or we
• 8:00
  could do it ourselves.
• 8:01
  Or we could put together an internal team that will help us do an AI transformation.
• 8:05
  And I think you're gonna see more companies say, the answer is in the room already.
• 8:09
  What we really need is a team of people who understand our context, our workflow, and who
• 8:13
  can make the gains of this AI productivity extend beyond just if if it only lives in
• 8:18
  engineering, unless you're a software development firm, there won't be enough ROI for
• 8:23
  this stuff to pencil.
• 8:24
  It has to get into finance, ops, customer support.
• 8:27
  It has to get into all these other areas for at the organizational level for the ROI to make
• 8:31
  sense.
• 8:31
  Yeah.
• 8:32
  And for Sierra, it's pretty obvious because they have that technical talent that lives
• 8:35
  within the organization.
• 8:36
  But would you say that's the first phase?
• 8:37
  It's taking a good look at your organization and having an honest conversation.
• 8:41
  Do I have the talent that lives within this to run that experiment, or should I outsource
• 8:44
  that?
• 8:45
  I think though we're also at interesting stage where because the tools are relatively
• 8:50
  accessible, I'd argue what you really need are people who are curious and engaged with the
• 8:55
  tech.
• 8:56
  If there are people who are curious and engaged with the tech, they can come up with lots of
• 8:59
  interesting ideas.
• 9:00
  It doesn't have to be just the most senior people in your org.
• 9:02
  It doesn't actually have to be your most senior people leading this transformation.
• 9:06
  What you really need are the people who understand what the tools can do.
• 9:09
  I think the SaaSpocalypse people are gonna look back and say, directionally, were right.
• 9:14
  But in the end, I think people are gonna use headless versions of systems of record because
• 9:18
  those systems are very good at keeping track of things.
• 9:21
  But you won't interact with the UI and you won't use them in the same way, but they're
• 9:25
  still gonna be a part of the tech stack.
• 9:26
  They're still gonna be very valuable important systems.
• 9:28
  That was one of the most notable takeaways that I felt like was buried in the piece.
• 9:32
  Neil said, quote, we found it's best to work with your systems of record, not replace them.
• 9:37
  GitHub keeps the PR.
• 9:38
  Salesforce keeps the account, and Linear keeps the issue.
• 9:40
  The agent is a layer across them.
• 9:42
  So it's just Yeah.
• 9:42
  Essentially a big pushback on SaaS's dead argument.
• 9:45
  Well, our next good read is another post on X from Everett Randall at Benchmark called the
• 9:50
  task economy.
• 9:52
  Everett opens by contrasting two metrics, tokens, which measure how much a model gets
• 9:57
  run, and tasks, which measure whether a model is actually getting better at real work.
• 10:02
  Tokens have obviously become the default way everyone tracks AI's growth, but he argues that
• 10:06
  just telling your usage is up, not that the model is improving.
• 10:09
  His core argument is that the Internet is basically tapped out as the training source.
• 10:14
  So the next real unlock for AI has to come from somewhere else, the knowledge sitting in
• 10:18
  professionals' heads that never got written down anywhere online.
• 10:21
  He estimates 99 of the expertise we actually want AI to learn, how a lawyer reasons through
• 10:26
  a contract, how a doctor works a diagnosis, lives only in people's heads.
• 10:30
  It's not on the Internet.
• 10:32
  And his answer is what he calls the task economy, paying real experts to attempt actual
• 10:36
  work tasks and then grading or correcting their output so that judgment gets fed back into the
• 10:41
  model as training data.
• 10:42
  He points to an impressive growth numbers as proof that this is already happening at scale,
• 10:46
  labs increasing their data spend 10 x year over year, and Benchmark's own portfolio company,
• 10:51
  Merkor, going from 1,000,000,000 to 2,000,000,000 in ARR in just four months.
• 10:55
  Charles, do you like the task metric better than token usage?
• 10:58
  I assume yes, but wanted to hear what you think limitations could be on this one or how it
• 11:01
  could be gamified.
• 11:02
  I think when people told me about, oh, all of this AI stuff is gonna be outcome based
• 11:07
  pricing.
• 11:07
  I'm like, well, what's an outcome?
• 11:08
  And they're like, what do you mean?
• 11:09
  I go, well, some things are easier to quantify like a support ticket.
• 11:13
  When it's marked closed and resolved, that's that's an outcome.
• 11:17
  But if you're closing the books for the quarter, there's a million little micro
• 11:21
  outcomes that roll up to the, like, closing the so I was I was like, look.
• 11:24
  For outcome pricing to work, you have to have a clear, easy to measure outcome.
• 11:28
  I don't hear people talking about outcome based pricing as much as they did maybe a year ago,
• 11:32
  much more usage based pricing.
• 11:33
  You know, you look at, like, Handshake and some of these businesses that were not in the data
• 11:37
  labeling business that now are, they're ripping, which tells me that the models have an
• 11:41
  almost insatiable demand for trained labeled data.
• 11:45
  And it reminds me a lot of some of the things that Sarah Guo brought up in her post about
• 11:48
  untrainable, which I think we have a lot of things that are in people's heads that are not
• 11:52
  yet in these models.
• 11:53
  And to our earlier conversation in this episode, the question is, like, how willing
• 11:56
  will people be to part with that knowledge?
• 11:58
  Because you can kind of only get paid for it once.
• 12:01
  At some point, you've contributed to the model, the model knows, and you you can't resell that
• 12:05
  knowledge or insight again.
• 12:07
  And so I think if I were a model provider, I'd be doing the same thing.
• 12:10
  This is probably an ongoing expense for you to capture real time, real world training data on
• 12:15
  these tasks.
• 12:16
  But my guess is that, like, in a year, it won't be 99%.
• 12:20
  It might be 80%.
• 12:21
  It might be 60%.
• 12:22
  This is, like, a one way transfer of knowledge from humans to machines.
• 12:25
  And I think there's money you may when I first heard about this market, I was skeptical that
• 12:29
  it had longevity.
• 12:30
  I'm like, isn't this just arbitrage?
• 12:31
  Don't you eventually, in a fairly short period of time, transfer all of that human knowledge
• 12:35
  to the machines, and then this is like isn't a thing anymore?
• 12:38
  And I don't know that I believe that anymore.
• 12:39
  Yeah.
• 12:40
  But that was my initial thought was like, this is a one time, one way transfer.
• 12:43
  How is this like really a thing?
• 12:45
  And it's not one time transferring your brain just because you continuously need to update,
• 12:50
  improve the models themselves so that it's actually ingesting the things that your brain
• 12:54
  would actually wanna put on paper.
• 12:55
  Right?
• 12:55
  I think at some point, law will be solved, for example.
• 12:58
  But then we'll have to say, well, what does a physicist do?
• 13:00
  Or what does a biologist do?
• 13:02
  Or what does a air traffic controller do?
• 13:04
  Or what does a an architect?
• 13:05
  We will move across domains, and there will be new datasets to label and new information to
• 13:12
  contribute.
• 13:12
  But I think it will be like locust to some extent.
• 13:15
  We will come and destroy all of the, like, offline knowledge in a given field.
• 13:19
  We'll get enough of it into the models that they're good enough, and then people will go
• 13:22
  say, well, what's the next area of expertise that we should go mine and acquire?
• 13:26
  Someone down in the threads also made the point that the task economy can come in the form of
• 13:30
  AI creating tasks for humans to do.
• 13:33
  Yeah.
• 13:33
  So like Uber fares or DoorDash tasks, which calculate the value based on how complex each
• 13:40
  task is.
• 13:41
  So the argument from their standpoint was if AI is now just selling work, then surely the
• 13:45
  previous ways of how you bill it has to change.
• 13:48
  That I think those are really specific examples because those are total gig economy works.
• 13:53
  We've mostly been talking in the examples of white collar professionals.
• 13:56
  Yeah.
• 13:56
  I mean, people keep saying it's gonna upend consulting.
• 13:58
  I'm like, maybe.
• 13:59
  Maybe not.
• 14:00
  I don't know.
• 14:00
  I think the consultants are gonna have something to say about that too.
• 14:03
  And, you know, every pricing change has all these weird dynamics.
• 14:07
  And I don't know.
• 14:07
  My experience has been, I spend a lot more on software than I used to.
• 14:12
  And even though software has gotten cheaper, I just use it more and buy more of it.
• 14:16
  So I have a lot of questions right now about does all of the math on this stuff pencil in
• 14:20
  the end?
• 14:21
  Do you think there's a world in which white collar professional people are doing this
• 14:25
  purposely as a side gig in putting some of their knowledge know how into the models
• 14:29
  themselves, or are they just doing it passively just by using the models themselves day to day?
• 14:33
  I think there are certainly people who are doing it as a side hustle, a GLG style expert
• 14:39
  network type.
• 14:39
  This is a way to make some money on the side.
• 14:41
  And other people, I think, maybe don't realize that they're helping to train these models by
• 14:44
  asking questions and contributing data.
• 14:46
  As they get better, we're just gonna use them more.
• 14:48
  Well, we have a fourth Goodread almost as a mini news flash.
• 14:51
  New York this week became the first state to fully pause new large data center construction
• 14:56
  imposing a one year moratorium on any facility using 50 megawatts or more of power.
• 15:01
  Governor Kathy Hochul cited rising utility bills, strained water supplies, and pressure on
• 15:06
  local communities as the reasoning, and she's also pushing to repeal the sales tax exemption
• 15:10
  hyperscalers currently get on these facilities.
• 15:13
  During the pause, the state's environmental regulator won't issue new permits while
• 15:16
  officials work out consistent standards for how data centers get reviewed and built going
• 15:20
  forward.
• 15:21
  New York already has more than 12 gigawatts of large power hungry projects, including data
• 15:25
  centers waiting to connect to its grid.
• 15:26
  And nationally, only about a third of Americans say they actually support the pace of data
• 15:31
  center construction happening around them.
• 15:33
  Wanted to just get your reactions if you think this type of legislature will actually do
• 15:37
  anything.
• 15:37
  I think a lot of the job of a politician is to read public sentiment and to pass legislation
• 15:43
  that addresses the needs of their constituents.
• 15:46
  I think tech is totally losing the communications battle.
• 15:50
  I won't even go into the data around the claims on energy and water because your mileage may
• 15:55
  vary depending on where you live.
• 15:57
  I don't wanna go down that road.
• 15:58
  Tech has totally lost the narrative battle here.
• 16:01
  And I think people say, like, well, what happens when you live in a bubble?
• 16:04
  I think if you live in the tech world that we live in, this AI feels mostly good.
• 16:08
  It's probably not gonna take your job.
• 16:10
  It's making your job more fun and interesting.
• 16:12
  But for a of other people, they're like, what's in it for me?
• 16:15
  I use the free Planet Chat GPT.
• 16:17
  Hasn't changed my life.
• 16:18
  I see this big ugly structure being built in my backyard.
• 16:21
  My electric bills up, whether that's due to inflation or due to the presence of these these
• 16:26
  things.
• 16:26
  I'm reading things online about the water consumption, whether and then some of that is
• 16:30
  motivated reasoning.
• 16:31
  And I think if you ask the average version, what's in it for me?
• 16:33
  I don't know what I would tell them.
• 16:34
  Why you should have a data center in your backyard or why, like
• 16:37
  And the new jobs thing, I think, isn't landing.
• 16:40
  Like, I don't think people are like, oh, I and I wanna work in a data center.
• 16:43
  Yeah.
• 16:43
  I wanna be a construction guy pulling cable or I think because within tech, AI has felt so
• 16:48
  inevitable and so magical, I don't think we've asked enough questions about why are so many
• 16:54
  other people so uninterested or act again, we talked about this before.
• 16:58
  This is the first tech wave in my career where I've seen active organized opposition.
• 17:03
  Other things people have been like, I don't know.
• 17:04
  I'm neutral on mobile or mobile's good.
• 17:06
  And maybe you could argue social media in its twilight era, people have come to see it
• 17:10
  through a different light.
• 17:12
  This is people are trying to stop something from happening as opposed to undoing something
• 17:16
  that's already happened.
• 17:17
  And I think tech needs to take these things very seriously and ask ourselves a question,
• 17:22
  what is in it for the average person?
• 17:23
  And it's funny, like, some of the AI people are just like, your job's going away.
• 17:26
  I'm like, well, that's also not helping the narrative, but it also might be true.
• 17:30
  So we have this interesting tension between what is the narrative that we wanna tell people
• 17:34
  about You is good and what's the truth in real and we really don't know what's gonna happen to
• 17:37
  Jobs.
• 17:38
  Saying that you shouldn't worry feels like a very privileged argument.
• 17:41
  And do you think the communications battle being so poorly lost by tech is a result of
• 17:47
  them just moving too quickly and not thinking it's important, or just assuming those people
• 17:51
  just don't have any voices or skin in the game to matter?
• 17:55
  I think most people in tech view AI advances as inevitable.
• 17:59
  So trying to slow them down or stop them feels like very decel and like a waste of time.
• 18:04
  And then instead of engaging with people about their real concerns, they're just like, don't
• 18:07
  be a Luddite.
• 18:08
  Get over it.
• 18:09
  This is gonna happen.
• 18:10
  That's not how you win people over to an argument.
• 18:12
  That's makes people feel like you just don't care about them.
• 18:14
  And I don't hear many people addressing the needs of these folks.
• 18:19
  Some of these data centers are like, they're off the grid.
• 18:21
  They are like separately powered by other things.
• 18:23
  They're not taxing the grid, but that's like a nuanced argument.
• 18:26
  I just feel like tech is not engaged with people on the issues that they seem to care
• 18:30
  about, which are these things are big.
• 18:32
  They consume a lot of power, and they are very visible.
• 18:35
  You know, AI is very abstract.
• 18:37
  A data center is a physical representation of this abstract fear that I think people have.
• 18:42
  And seeing three of them go up in your community makes you think, oh, the walls are
• 18:46
  literally closing in on me.
• 18:47
  This future state that I'm afraid of is happening even faster than I thought.
• 18:50
  Well, we will certainly see more discourse about that over the weekend, but we are all out
• 18:54
  of time today.
• 18:54
  Thank you, Charles, for stopping in.
• 18:56
  Thank you to our listeners for hopping in as well.
• 18:58
  Please like and subscribe, and we will talk with you all next week.
• 19:00
  Thank you.
• 19:01
  Bye.