Latest / The Learning Corner by Precursor / Episode #47: We Are The Enemy, The AI Productivity Paradox, Anthropic's Author Class Action

Transcript

Episode #47: We Are The Enemy, The AI Productivity Paradox, Anthropic's Author Class Action · The Learning Corner by Precursor

• 0:00
  Think we've gone through this.
• 0:01
  I don't know if it's 12 stages of grief or whatever you wanna describe it.
• 0:04
  We've definitely have gone from denial.
• 0:05
  Big funds aren't a problem.
• 0:06
  To acceptance.
• 0:08
  Big funds are having an impact on seed.
• 0:10
  To this new world where people are trying to figure out, if this true, what do I do?
• 0:14
  It's the week of September 8.
• 0:16
  Welcome to the Learning Corner, a Precursor Ventures podcast.
• 0:24
  Welcome to the Learning Corner by Precursor.
• 0:26
  I'm Mia Farnam, I'm joined by RGP and managing partner Charles Hudson, where we will be
• 0:31
  talking through the articles and Goodreads that caught our teens' attention over the past week.
• 0:36
  Our first topic this week is a new blog post from Euclid Ventures called We Have Met the
• 0:41
  Enemy, and He Is Us.
• 0:43
  Euclid Ventures is also on the leaderboard for top people we feature here.
• 0:46
  I felt like the piece was a wonderful blend of so many podcast episodes we've done over the
• 0:52
  last few months with a large takeaway being that venture capital has drifted away from its
• 0:56
  roots as a market for independent thinkers.
• 0:59
  As we've mentioned many times before, mega funds and small number of startups are now
• 1:03
  attracting a large share of the capital.
• 1:05
  Remember, nearly 40% of all USBC dollars last year went to just 10 companies.
• 1:11
  This level of concentration has created a consensus driven environment where many people
• 1:15
  pursue the same deals, reducing differentiation and limiting genuine discovery.
• 1:20
  The piece also highlights how the influence of mega funds extends beyond returns and into the
• 1:24
  broader early stage ecosystem.
• 1:26
  Large and small fund alike are chasing an extreme version of the power law model,
• 1:31
  searching for multibillion dollar outcomes.
• 1:33
  The data, however, suggests that this approach rarely produces results and analyses show
• 1:38
  little correlation between deal price and success.
• 1:41
  And the share of startups advancing from c to series a has declined compared to previous
• 1:45
  years as we've mentioned.
• 1:46
  The takeaway is that consensus investing and copycat behavior may be weakening the health of
• 1:51
  the market rather than fostering innovation.
• 1:53
  If you had to give someone one piece to summarize the general vibes and questions
• 1:57
  bubbling up across early stage venture capital in today's market, I'd honestly choose this
• 2:02
  one.
• 2:02
  Anything else, Charles, that you felt the Euclid team highlighted here that hasn't been
• 2:06
  walked through in as much depth before?
• 2:08
  I think you can draw a through line from Rob Goh's piece that we talked about and what we
• 2:12
  talked about previously about consensus versus non consensus.
• 2:16
  The one thing I will say is I've noticed a shift in the we'll call them the seed
• 2:20
  specialist funds, funds like ours that don't have aspirations to become big multi stage
• 2:24
  players.
• 2:25
  I think we've gone through this I don't know if it's 12 stages of grief or whatever you wanna
• 2:28
  describe it.
• 2:29
  We've definitely have gone from denial, big funds aren't a problem to, like, acceptance big
• 2:34
  funds are having an impact on seed, to this new world where people are trying to figure out, if
• 2:38
  this is true, what do I do?
• 2:40
  And what does this mean for me now that the sort of the shock of capital concentration,
• 2:44
  both at the company level and at the fund level, is kind of an established reality.
• 2:48
  And a lot of fund managers I've talked to, in their more honest ones, have said, hey, I think
• 2:52
  I'm good at this job, but I'm not in OpenAI or Anthropic.
• 2:56
  Those aren't companies that are in in our portfolio, and we're not some huge mega fund.
• 3:01
  What do I still have the right to exist?
• 3:04
  And I think there's just a lot of soul searching I'm seeing going on, particularly
• 3:07
  amongst seed fund managers who are trying to figure out what does it mean to be a seed fund
• 3:11
  manager in this world?
• 3:12
  What role are we filling in the ecosystem?
• 3:14
  And I'd argue if the role you're filling in the ecosystem is slightly front running multi stage
• 3:18
  funds and the things that they will eventually get to, that feels like an arbitrage that will
• 3:23
  eventually get squeezed.
• 3:24
  And the irony is the further you try to get away from the shadow of the big funds and what
• 3:29
  they're interested in, what they fund, the less overlap you have and the harder it is to talk
• 3:32
  to limited partners and other people about why the stuff you're doing is interesting if it's
• 3:37
  not in the wheelhouse of those big funds.
• 3:38
  Yeah.
• 3:38
  Maybe an emerging fund manager might ask you, because you've done this for five funds, how do
• 3:42
  you thread the needle of something in the middle, it feels like, between those two
• 3:45
  strata?
• 3:46
  I think the things I would tell them is you have to be right.
• 3:49
  So the things you've picked, they all ultimately have to be right.
• 3:52
  They can't just be different and interesting or just things you find fascinating but aren't in
• 3:56
  fact good businesses.
• 3:58
  So that's one.
• 3:58
  Two, I would say you have to have an LP base that understands your strategy and understands
• 4:02
  the timeline for your strategy.
• 4:04
  So if you're doing something that's like currently out of favor and will remain out of
• 4:07
  favor for a while, make sure you have LPs who have the stomach for that.
• 4:11
  If you have somebody that's out of favor but you feel like it's on the cusp of an inflection
• 4:14
  point, that's a really different story.
• 4:15
  I think it's it's emotionally difficult to sort of sit out the party or to do other things and
• 4:21
  have people tell you that you're either wasting your time or the things that you're working on
• 4:24
  are dumb.
• 4:24
  That's not easy to hear day in and day out, and not everybody can do that.
• 4:27
  I think for some fund managers, they wonder if to your point of finding like minded LPs who
• 4:33
  understand the journey, if what you're looking for isn't multistage venture fund world.
• 4:37
  If they meet LPs who are already in a multistage fund, would they ever be interested
• 4:40
  in jumping into something that doesn't look like that?
• 4:42
  I find there are some LPs I meet.
• 4:44
  They think of the big funds as the building blocks of their portfolio, their core anchor
• 4:48
  positions.
• 4:49
  And one of two things happens.
• 4:50
  They either get into the big funds they wanna be in, then they're like, great.
• 4:53
  We are done with venture capital.
• 4:54
  We are in the names we wanna be in.
• 4:56
  We don't actually need anything else.
• 4:59
  Whatever we're gonna get from venture, we'll get from this group of people.
• 5:01
  There's another set of folks who say, now that I have the foundation done, I can roll up my
• 5:05
  sleeves and do the stuff that's maybe more interesting.
• 5:07
  I can go find a sector specialist fund in some area that I'm interested in.
• 5:11
  I can go find a generalist fund that's stage specific that's doing something that I think is
• 5:15
  interesting.
• 5:15
  And I find people tend to hard bifurcate once they've got the base of their portfolio built.
• 5:20
  Well, speaking of big funds, our next topic this week is a post from Sequoia titled the AI
• 5:25
  productivity paradox, high adoption, low transformation.
• 5:29
  It digs into a question we've been discussing a lot here.
• 5:32
  If AI adoption is happening everywhere, why aren't companies seeing real lifts in
• 5:36
  productivity or bottom line results yet?
• 5:38
  The piece walked through both the MIT NADA report we chatted through last week in addition
• 5:42
  to Brynjolfsson's Canaries in the Coal Mine paper that highlights the progress and
• 5:47
  challenges posed by AI, specifically across the labor market and entry level roles.
• 5:51
  For founders building in a space, the big takeaway is that tools need to go way deeper
• 5:57
  than surface level output generation.
• 5:59
  The winners will be the ones designing AI products that actually integrate into messy
• 6:03
  real world workflows, adapt to how teams work, and solve specific unsexy business problems
• 6:08
  that live behind the scenes.
• 6:10
  It's a really useful reminder that we're still in the very early innings, and the gap between
• 6:14
  using AI and actually operationalizing it is where a lot of opportunity still lives.
• 6:19
  I have a few questions here, but wanted to pause for your high level opinions first.
• 6:23
  I read another article that was I think it might have been in Axios.
• 6:27
  It was the practice leader from Accenture who does enterprise AI transformation.
• 6:31
  She said, there's a difference between transforming enterprise and using AI.
• 6:35
  And I'm like, I don't know.
• 6:37
  Those who've listened to this podcast probably are used to my somewhat skeptical view of where
• 6:42
  we are in the adoption cycle.
• 6:43
  And in many ways, I feel like all of this failure and gnashing of teeth is a necessary
• 6:48
  part of transformation because if it was just as easy as giving everybody in your company
• 6:52
  ChatGPT and stepping back, this wouldn't be that revolutionary.
• 6:56
  The revolutionary work is gonna come in enterprise transformation, which requires not
• 7:00
  just giving everybody AI tools or not just doing AI show and tell every week, but actually
• 7:04
  digging in to workflows and processes that you have and reimagining them.
• 7:08
  And anyone who's sold to the enterprise knows that's not something you do overnight.
• 7:12
  So this idea that just because we had the ChatGeePeeTee moment that in a year, these big
• 7:17
  lumbering enterprises were gonna magically transform to take advantage of AI, think was
• 7:22
  always a pipe dream.
• 7:23
  But it's a pipe dream in some ways that has fueled all this enthusiasm because I think the
• 7:28
  CEOs that I talked to of large organizations, they know that they have to get this right
• 7:33
  eventually.
• 7:34
  And just because these initial attempts have failed doesn't mean they're gonna quit.
• 7:38
  They're gonna try to learn from their mistakes.
• 7:39
  And I think it could be the back half of this decade, which I know is a lot longer than my
• 7:44
  peers think.
• 7:45
  The back half of this decade before people can really, really point to this stuff working
• 7:50
  across their organization in ways that are easy to imagine but hard to see today.
• 7:55
  It reminds me of back in consulting days when you would do Workday, SAP, Oracle
• 7:59
  implementations.
• 8:01
  First of all, the implementation would take half a year, and then the entire process after
• 8:05
  that, the post transformation, was extremely painful.
• 8:07
  It was a whole two year, three year engagement.
• 8:10
  Consulting teams were there trying to manage comms, handhold people, and that was for
• 8:14
  something that was a little bit more tangible than most AI tools.
• 8:17
  Today, it is funny that people assume that there's going to be so much value extracted
• 8:21
  immediately when, obviously, to your point, it's going to take so much longer.
• 8:26
  And also there's people involved.
• 8:27
  And different people have different levels of enthusiasm and adoption for AI.
• 8:32
  But I think within enterprises, very few things happen without coordination.
• 8:36
  So someone's gonna have to decide, this is what we're gonna do in finance with AI.
• 8:40
  These are the tools we're gonna use.
• 8:41
  These are the processes we're gonna automate.
• 8:43
  And here's how we're gonna measure success.
• 8:45
  And I think we haven't had that yet before.
• 8:47
  It's been like, let's just throw AI at finance and like, let's see what the finance team comes
• 8:51
  up with.
• 8:52
  That's not gonna work in most cases.
• 8:53
  Eric, who's the author of Canaries in the Coal Mine piece, he used a j curve graph to describe
• 8:58
  his productivity paradox back in the nineties and how AI will truly transform the labor
• 9:03
  market over time.
• 9:04
  The graph shows how an introduction of new tech initially causes a decline in productivity
• 9:09
  before organizations adapt to be able to capture the productivity benefit.
• 9:13
  This is very reminiscent of computers in the eighties and nineties, and its prediction is
• 9:17
  that AI will likely take a similar case.
• 9:19
  So what does this mean for investors and founders building today?
• 9:22
  I know that you think the time horizon's a little bit longer for value add, but
• 9:26
  Well, the real question is, is this gonna be a linear change or a step change?
• 9:30
  So if it's a linear change where, like, everyone's just gonna get a little bit better
• 9:34
  until they're good, then if you're an AI startup today and you have a long runway, you
• 9:39
  can probably just use that long runway to kind of ride that curve.
• 9:43
  But if it's discontinuous, this is hypothetical, at the 2026, everybody says,
• 9:47
  we're just scrapping everything we've done and started and we're starting over.
• 9:50
  Now that we've played with v one, we're gonna push reset on all the projects we tried and do
• 9:55
  them better.
• 9:55
  In the same way that the first time people built a website, it kinda sucked.
• 9:58
  Yeah.
• 9:59
  And they're like, oh, now that we know what our we want our website to do, we're gonna actually
• 10:01
  build the website that we want, not the website we thought we wanted.
• 10:04
  And you start over, but it means throwing away a lot of the stuff you did before because the
• 10:09
  new capabilities are so much better.
• 10:10
  If that's the case, then having a huge balance sheet isn't actually as useful unless you can
• 10:16
  use that balance sheet to stay nimble when that big reset happens.
• 10:19
  I suspect it won't be a huge reset.
• 10:21
  But I don't know.
• 10:22
  I I spent two days teaching myself how to use cursor and Claude code and building some things
• 10:27
  and I totally got lost in coding.
• 10:29
  And I will say this, it broadened my horizons for other things I could build.
• 10:34
  And I think what we're gonna have are these iteration loops where people try things,
• 10:38
  imagine the future, try to build the future, they fail.
• 10:40
  They get closer that allows them to imagine an even more distant future.
• 10:44
  And it's just gonna take time.
• 10:46
  It's just gonna take time, which I know no investor wants to hear.
• 10:48
  I think everybody wants to think this generation of companies is the set of companies
• 10:52
  that matters.
• 10:53
  We thought that about companies in the nineties too, and we were right in some cases and wrong
• 10:57
  in others.
• 10:58
  Our last topic is an article by Reuters that shared the news that Anthropic agreed to pay
• 11:03
  $1,500,000,000 to settle an ongoing author class action suit filed last year where a group
• 11:09
  of authors sued Anthropic for using their work to train Claude models without their
• 11:13
  permission.
• 11:14
  Lawyers representing the author said that this was a landmark event as it was the largest
• 11:17
  copyright recovery in history and the first of its kind in this AI era.
• 11:22
  While the judge ruled that Anthropic made fair use of these authors' work to train Claude
• 11:26
  earlier in the summer, the millions of pirated copies of books stored in a central library
• 11:31
  were deemed a violation of rights.
• 11:33
  Anthropic and other foundational model companies' response is that they make fair use
• 11:38
  of copyrighted material to generate new innovative content.
• 11:41
  A trial was scheduled to begin before 2026 to determine how much Anthropic owed with
• 11:46
  potential damages ranging into the hundreds of billions of dollars.
• 11:50
  The debate over quote unquote fair use is pretty confusing to me and largely up for legal
• 11:57
  interpretation with very little precedent to guide judges.
• 12:01
  Any thoughts from your end on this one?
• 12:03
  You know, I think it's it is interesting.
• 12:05
  I think about fair use with YouTube and with music and all of these other domains.
• 12:10
  I think the difference is the Internet is just a lot bigger now and LLMs are a lot bigger than
• 12:17
  the products that we were arguing about this over before.
• 12:19
  So if you think that Anthropic and ChatGPT are going to be people's primary interface to
• 12:25
  information, $1,600,000,000 is not a lot of money for content that will be accessed and
• 12:31
  viewed by hundreds of millions or billions of people on a regular basis.
• 12:36
  And in many of these things, like once you've settled, unless there's some ongoing revenue
• 12:40
  share or some economic arrangement, it's a pay once license Yeah.
• 12:43
  Deal.
• 12:43
  And and that that to me doesn't feel like a lot of money if that's the end state.
• 12:47
  At the same time, I think the desire to stay on the cutting edge of training these models by
• 12:51
  getting your hands out as much information as you can creates, certainly in my mind, creates
• 12:56
  a temptation to do things like this.
• 12:57
  I don't know enough about the specifics of the case to know what what's being alleged.
• 13:01
  I've I've read through it, I'm I'm not a lawyer nor will I ever be one.
• 13:05
  But I I do think there are issues here.
• 13:07
  You know, early on we had this with actors' voices and and naming like this with things
• 13:11
  like Eleven Labs.
• 13:12
  Like, hey, once you have my voice and you can clone it and use it forever, like, what's that
• 13:17
  worth?
• 13:18
  Right.
• 13:18
  I think we don't know.
• 13:19
  And similarly, like, what are the ways that you can get around copyright that still thread this
• 13:24
  needle of being legal?
• 13:26
  Like, I remember, maybe it was Chatuche BT, Scarlett Johansson's voice.
• 13:28
  And she came up and, that's my voice.
• 13:30
  And they're like, no, it's not.
• 13:32
  I think, like, they're these are only going to continue to happen at more volume.
• 13:38
  And I do actually wonder, and I and I don't really know how they do this, like, the authors
• 13:41
  who were able to kind of figure out that these models were training unpirated versions of
• 13:45
  their material, was it just them using the tools and realizing that I didn't give rights
• 13:50
  for this?
• 13:50
  Because if that's the case, it'll take you to represent yourself as an artist in order to
• 13:54
  kind of like build the pace and speed for a case like this.
• 13:57
  That seems like a lot of work to gain that sort of momentum.
• 14:01
  And I mean, on the Internet, we figured this out eventually.
• 14:03
  You know, there were do not crawl.
• 14:05
  Like, there were ways to say, hey, leave my content alone.
• 14:08
  There were paywall.
• 14:09
  But I think the interesting thing is with agents, a lot of the things that we built in
• 14:13
  the internet, like, this is a site that's protected by a password.
• 14:15
  Well, an agent could navigate that fairly simply.
• 14:17
  It's like not that difficult.
• 14:19
  And I just think the stakes are so high right now.
• 14:22
  I'll be very curious to see what happens.
• 14:25
  My sense is that the appetite for large language models to get their hands out as much
• 14:30
  training data and content as possible will be the driving force of strategy.
• 14:34
  And it means that at the margin, I think it's a little bit of a weird prisoner's dilemma.
• 14:38
  You're like, what if I follow the rule and this other LLM cheats?
• 14:41
  This is like a total non sequitur.
• 14:43
  The one thing I've learned watching the Internet, it's an unfortunate thing.
• 14:47
  If you're a large company like Google or Microsoft, it is often better to infringe and
• 14:53
  get the benefit.
• 14:54
  Whatever penalties Google is ultimately assessed for whatever they're accused of doing
• 14:58
  with Chrome, you think about that, you're like, well, you're Google, I hate to say it, do you
• 15:01
  think you're better off having Chrome as a dominant browser and paying a fine ten or
• 15:04
  fifteen years later.
• 15:05
  Same thing with Microsoft.
• 15:07
  So I think the history has been that in these competitive network effect driven or winner
• 15:11
  take all markets, it is probably a better business strategy to win now and pay later than
• 15:16
  it is to actually stick to the rules.
• 15:19
  And I think there's a set of regulators who understand that, and I think there's a set of
• 15:22
  companies who understand that, and we'll see how that tug of war plays out.
• 15:25
  And even the folks who aren't winning now, paying later, I think they're probably
• 15:29
  experimenting with ways to wiggle around what's legal, what's not.
• 15:33
  We're big fans of tennis in my household.
• 15:36
  And when the number one tennis player, Janik Sinner, he got dinged for using illegal
• 15:40
  substances from a massage lotion, he didn't get penalized that hard.
• 15:45
  We have close friends who play in challenger tournaments.
• 15:47
  They're also on the ATP tour.
• 15:48
  And they mentioned that all the top 10 tennis players are constantly, every year, working
• 15:53
  with the best people in the business to try substances that aren't illegal today, but they
• 15:56
  are illegal next year.
• 15:58
  And so basically, every year they revamp, but they have an edge over everyone else because
• 16:02
  they have the money and the pace and the access to get there.
• 16:05
  So it's not illegal today, but it will be soon.
• 16:08
  Did that mean that it's allowed?
• 16:09
  If you're Janik Sinner, don't come for me.
• 16:11
  Your routine alcraz at my house.
• 16:13
  One last question here.
• 16:15
  If you're a founder building out the application layer in the creative or artistic
• 16:18
  space, especially where copyright might come into play, what should you be prioritizing
• 16:23
  right now if you have control?
• 16:24
  I I think it's really interesting.
• 16:26
  If you look at the history of even Eleven Labs, they kinda bumped their toe, I guess is the
• 16:31
  best way to describe it, with the sort of the Hollywood copyright universe.
• 16:36
  That's a set of folks who have pretty successfully battled almost every technology
• 16:40
  wave and and managed to extract something of value for the content that they and there was a
• 16:46
  period of time where I think people thought, gee, maybe what eleven Labs maybe that Bubba's
• 16:49
  slumping their toe will be the end of that company, and it's very much not been the case.
• 16:54
  Mhmm.
• 16:54
  And so I think if you're gonna be in the creative field, I think you have to have a plan
• 16:58
  for how are you gonna engage with rights holders and license holders.
• 17:02
  And in some cases, there's the Napster strategy, is you try to get big enough and
• 17:07
  force them to deal with you on your terms.
• 17:08
  That can be tough.
• 17:09
  There could be the Spotify strategy, which is you engage with them sort of on their own
• 17:13
  terms, but try to make the deal work for you as well, but you kind of acknowledge where they're
• 17:17
  coming from.
• 17:17
  There's a bunch of different models here, but I think this is gonna be highly fraught.
• 17:21
  And some of these things go away.
• 17:22
  I remember three years ago, everyone's like, Gen AI and art.
• 17:26
  This is like the worst thing and that the the battlefield seems to have shifted to Yeah.
• 17:29
  Other domains.
• 17:30
  Not saying that it's been settled.
• 17:32
  It's just like the energy shifted to a different place.
• 17:33
  Well,
• 17:34
  we are all out of time today.
• 17:35
  Thank you, Charles, for calling in.
• 17:37
  Thank you to our listeners for stopping by.
• 17:39
  Please like and subscribe, and we will talk with you all next week.
• 17:41
  Thank you.
• 17:42
  Bye.