YT YouTube2Text

YouTube2Text → Transcripts → Palantir CEO Says Only 2 Types of Jobs Are SAFE From AI

Palantir CEO Says Only 2 Types of Jobs Are SAFE From AI — Transcript

by Finance Bureau · 2,511 words · 384 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Low-end code, routine contracts,
0:02standard [music] copy. This is the sort
0:05of work that once trained new graduates
0:07entering the workforce, [music]
0:09but now it all runs on a quick prompt.
0:12In offices around the world, the desks
0:14[music] are still there, but the work
0:16that justified them is starting to
0:18leave. And everyone who spent their
0:20early 20s and a lot of money preparing
0:23for those jobs is now wondering
0:28what work is a actually Alex [music]
0:31Karp has an answer. Only two groups of
0:34people will really thrive [music] in the
0:36coming age of AI, and neither of them is
0:39the typical graduate that the system
0:41[music] has been producing. So, to
0:43understand his argument, we'll look at
0:45the work that's going first, the case
0:47for the two groups he backs, [music] and
0:49the company that's already reorganized
0:52its hiring around this whole concept. My
0:55name is Guy, and this [music]
0:56is the Finance Bureau.
1:00Now, Alex Karp is known for being, shall
1:03we say, pretty forthright with his
1:05opinions. And back in March, on the tech
1:08show TBPN, he laid out his views on
1:11where he reckons AI is going without
1:14mincing his words. Quote, "There are
1:16basically two ways to know you have a
1:18future. One, you have some vocational
1:21training, or two, you're neuro
1:23divergent." Not really a comforting
1:25outlook on the idea of humans and
1:27machines working side by side. Now, as
1:29Karp sees it, only two groups stand to
1:32have any kind of certainty in the future
1:34ahead. So then, who exactly are these
1:37two groups? Well, the first is
1:39essentially the people who work with
1:41their hands. So, think electricians,
1:44plumbers, carpenters, welders, etc. The
1:46people you call when something breaks or
1:48needs fitting, and ironically, the
1:51trades that plenty of schools spent a
1:53long time steering bright students away
1:56from. The second group is people whose
1:58brains are wired differently. Karp uses
2:02the term neurodivergent, and in case
2:04you're unsure about the meaning there,
2:06that's an umbrella term for conditions
2:08like dyslexia, ADHD, and autism, where
2:11the brain processes information
2:13differently from a typical pattern. In
2:15other words, the people the standard
2:18classroom was least designed for. Now,
2:21there's an amusing bit of detail about
2:23who's delivering this message. Karp
2:25studied at Haverford College in
2:27Pennsylvania, took a law degree at
2:29Stanford, and then earned a doctorate in
2:31social theory from Goethe University in
2:33Frankfurt. So, something of an academic.
2:37But, bear in mind his case isn't against
2:40thinking. He's making a case against a
2:42particular kind of education and the
2:45particular kind of job it was made to
2:48prepare people for, which means before
2:50we go any further, we need to look at
2:53what's being left behind. So, what
2:56exactly is AI taking first? Well, not
3:00the top of the profession, but the
3:02bottom rungs instead. Consider for a
3:04moment how a career in software used to
3:07begin. A graduate entered a tech firm
3:09and got handed the jobs nobody senior
3:11wanted, like first-pass code and simple
3:15debugging. The small fixes that piled up
3:17in the ticket queue. It was tedious, but
3:20it was also how you learned, because
3:22every bug you chased taught you a little
3:25more about how the whole system fitted
3:27together. That work is now the first
3:30thing an AI coding assistant does. The
3:33legal profession is in a similar
3:35position. A first-year associate spent
3:37their early years on routine contracts,
3:40reading clause after clause until they
3:43could spot the one that would cause
3:44trouble. Now, firms are trimming their
3:47junior intake because an AI model can do
3:50the first read. And copywriting went
3:52down the same road as well. The product
3:54descriptions and standard emails that
3:56taught a young writer how to how to hit
3:57a brief have moved into automated
4:00workflows. So, it's fairly easy to see
4:03how things go from here. The senior
4:05engineers, partners, and editors are
4:07safe for now, at any rate, but the
4:09junior work that trained them is
4:11disappearing. And that leaves confusion
4:14and uncertainty lingering in every
4:16office because if the junior work is
4:19gone, how does anybody become senior?
4:22The judgment you need to check what an
4:24AI produces was always developed by
4:27doing the dull work yourself over and
4:30over for years. Take that away and the
4:33apprenticeship goes with it. Now, Karp's
4:35argument goes one step further than the
4:37jobs themselves. He goes after the
4:40education system that typically precedes
4:42them because in his view, a lot of what
4:45universities sell is compliance. Learn
4:48the framework, apply the framework, and
4:51then produce the answer that's expected.
4:53But when you look at it that way, it
4:55pretty much sounds like work that a
4:57machine tends to do very well. Now, Karp
5:00has been especially harsh on the
5:02humanities and standard philosophy
5:04tracks, warning that AI will gut them
5:07along with the corporate jobs they were
5:09made to supply. He said that elite
5:12generalized degrees, especially in the
5:14humanities, will be hard to sell without
5:16a specific skill alongside them because
5:19a degree that teaches you to follow a
5:21specific system is training you for a
5:24job that machines are about to take. So,
5:27while graduates may not be short of
5:29talent, the trouble is they were trained
5:31to follow processes that software now
5:34handles better. Not exactly the best
5:36return on years of tuition. And with AI
5:40now reshaping who gets hired, where, and
5:43on what terms, it's clear that the world
5:45is going to change a lot in the coming
5:48years, which is exactly why we started
5:50the Finance Bureau newsletter. It's
5:52completely free, and every week we break
5:54down the biggest developments in the AI
5:56buildout, the industries it's
5:58transforming, and everything else
6:00happening across the markets. So, just
6:02click the link in the description or
6:04scan this QR code on your screen to get
6:06started. Right, let's get back to Carp's
6:09first group, the people with practical
6:11skills that AI is nowhere near
6:14replacing. Consider what it is about a
6:16trade that software can't touch. Imagine
6:19an old house with a flooded basement.
6:22Nobody knows where the water is coming
6:24from and there are no plans showing
6:26where the pipes run. A plumber arrives
6:28and starts inspecting the old cast iron
6:31pipes, tapping them with a wrench and
6:33listening for signs of damage.
6:35Eventually, they find a cracked section
6:37that's been leaking for God knows how
6:39long and the damaged pipe has to be
6:41removed without disturbing the rest
6:43while a replacement fitted at exactly
6:46the right angle is needed to keep the
6:48water flowing. Get it wrong and the
6:51basement could flood all over again and
6:54throughout the process, everything has
6:56to meet local building codes for good
6:58measure. Now, consider how many
7:00decisions just got made in that
7:03basement. Which pipe to trust, what a
7:05different tone meant when a particular
7:07piece of pipe got tapped, how much of an
7:09old system can be disturbed before
7:11something else gives way. All of that
7:13knowledge came from someone standing in
7:15the water and reading the room with
7:18their collected senses and experience.
7:20And there's a name for why that's so
7:23hard to automate. It's called Moravec's
7:26paradox after the roboticist Hans
7:29Moravec. Basically, the things we think
7:31of as difficult like logic, calculation,
7:34and writing code turn out to be cheap
7:37and relatively easy for computers. The
7:39things we barely think about though like
7:41balance, touch, and working with an
7:43awkward tool in a cramped space, they
7:46turn out to be hugely expensive and
7:48difficult for computers. So, that's the
7:50heart of Carp's first group. The
7:52textbook knowledge, how a drain should
7:54fall or how a circuit should be wired,
7:57can be written down and a model can
7:59recite it. What doesn't transfer to
8:01software though is the judgment on site,
8:04the part when someone is physically
8:06there. And that's why the trades, long
8:09considered as the fallback option, are
8:11right at the top of Karp's list. But
8:13Karp's second group is perhaps less
8:16obvious, and it's the one he takes
8:18personally. Karp is dyslexic, and for
8:21someone with serious dyslexia, the
8:23standard route through school comes with
8:26a lot of friction. The textbooks and the
8:28worksheets, the step-by-step methods
8:30everyone else follows, don't really work
8:33the way they're supposed to for someone
8:35whose brain is wired differently. So,
8:37they're forced to find another way to
8:39the answer. And Karp credits that with
8:42how he thinks. In his words, quote, "If
8:45you are massively dyslexic, you cannot
8:48play a playbook, and therefore we learn
8:50to think freely." Now, in Karp's
8:53argument, a playbook is the agreed way
8:55of doing things. Think the template for
8:57a contract or the standard structure for
9:00a report. These are the approved methods
9:03for a problem that's been solved a
9:05million times before. Throughout its
9:07history, corporate work has been about
9:09running playbooks well, and education
9:12has been about teaching people to run
9:14them. But, a playbook is written down,
9:17and anything written down a large
9:19language model can learn. So then, if an
9:22AI runs the standard approach faster
9:25than any graduate, who has anything left
9:28to offer? Well, according to Karp, it's
9:30the people who never fitted the standard
9:33approach to begin with. The dyslexic
9:35thinker who learned to see a problem
9:36whole because reading it line by line
9:39wasn't really an option. The person with
9:41ADHD who's familiar with ambiguity,
9:44jumping between ideas a more orderly
9:46mind would never connect. Or the
9:48autistic engineer who spots the pattern
9:51or the anomaly that everyone else's
9:53process is likely to smooth over. For a
9:56long time, those traits were considered
9:58problems to be managed. They didn't fit
10:01the interview format, and they clashed
10:03with performance reviews that rewarded
10:05doing the expected thing reliably. But
10:08now, the expected thing is what AI has
10:12started to take over. So, thinking that
10:14was once a handicap in a world of
10:16standardized systems becomes the one
10:19thing that AI can't supply. And that's
10:22what ties Karp's two groups together,
10:24even if a welder and an autistic
10:26software engineer don't, on the face of
10:29it, appear to have much in common.
10:31Neither of them does work that can be
10:33boiled down to a set of instructions.
10:36One reads the physical world while the
10:38other reads problems sideways. And Karp
10:41hasn't just talked about this. He's put
10:43Palantir's hiring approach behind the
10:46idea. So, what does a company look like
10:49when it actually acts on this thinking?
10:51Well, for starters, it stops relying on
10:53the usual pipeline. You see,
10:55historically, elite firms have recruited
10:57the same way. Top university, top
11:00grades, a polished internship, a
11:02polished interview. Palantir has set up
11:05two doors that go around that route in
11:07entirely. And the first of them opened
11:10in April last year. It's called the
11:12Meritocracy Fellowship, and it's aimed
11:14at people who've just left high school.
11:17Not university graduates, mind you, but
11:19recent school leavers who aren't
11:21enrolled on a degree course and don't
11:24hold one. To get in, you need at least
11:261,460
11:28on the SAT or 33 on the ACT, plus some
11:33experience with programming. The
11:35fellowship runs for 4 to 5 months in New
11:37York and pays $5,400
11:40a month. That's not a bad deal,
11:42especially for people who, in the normal
11:44scheme of things, would be paying a
11:46university for the privilege of being
11:48there. Hundreds applied for a first
11:50cohort of only a couple of dozen places.
11:53And what happens once you're in is
11:56rather interesting because for the first
11:585 weeks, fellows don't get to touch a
12:01line of code. Instead, they study
12:03Western civilization, American history,
12:06philosophy, and religion from Frederick
12:08Douglass to a visit to Gettysburg.
12:10Somewhat surprising a curriculum for a
12:13company whose boss is so scathing about
12:15the humanities. But bear in mind what
12:18Karp's complaint is actually about. He
12:20has nothing against reading hard books
12:22and discussing them. But what he objects
12:24to is the compliance culture that's
12:27developed around them. After the
12:29seminars, the fellows go straight onto
12:31real Palantir engineering teams working
12:34with live clients. And the strongest of
12:37them can then go on to interview for
12:39full-time roles. Then Palantir has a
12:42second avenue. In December last year,
12:44shortly after a clip of Karp fidgeting
12:47on stage at the New York Times DealBook
12:49Summit went viral, Palantir announced
12:51its neurodivergent fellowship. And the
12:54company is clear that it isn't a
12:56diversity initiative. Palantir's
12:58position is that neurodivergent people
13:01will be the elite builders of the next
13:04technological era. And it's hiring for
13:06every kind of role on that basis. Now,
13:09no diagnosis is required to apply, but
13:12self-taught candidates are encouraged
13:14alongside formally educated ones at any
13:18stage of their career. The roles are
13:20full-time, based in New York or
13:21Washington, and pay between $110,000 and
13:25$200,000 a year.
13:27And the final round of interviews is
13:29conducted by Karp himself. So, the chief
13:33executive is personally sitting across
13:35from candidates the old system would
13:37have filtered out before they ever
13:39reached a recruiter. Now, reports vary,
13:41but the fellowship drew somewhere
13:43between 1,000 and 2,000 applications
13:46within days of launch. So, let's
13:48consider what Palantir has actually done
13:51here. It's hired school leavers with no
13:53degree and put them on live client work
13:56while also opening full-time roles to
13:58self-taught people with no credentials
14:01to show. And it's told neurodivergent
14:03applicants that the way they think is
14:06the reason it wants them, not something
14:08to be accommodated. Plus, on top of it
14:11all, its chief executive has made
14:13himself the final gatekeeper. Now,
14:16whatever you may think of Palantir, and
14:18safe to say many people aren't exactly
14:20fans of the company, it's nonetheless
14:22likely giving us an early glimpse of
14:25where the job market is heading. For
14:27most of modern history, the safe
14:29approach was to learn the standard way
14:31of doing things and do it better than
14:33the next person. But, as AI takes over
14:35more and more of that work, employers
14:38will place greater value on skills that
14:40machines can't easily replicate. Whether
14:43that's a plumber who can diagnose a
14:45problem no software could solve, or
14:48someone who thinks differently enough to
14:50find the solutions that nobody else
14:52considered. And who knows? Perhaps the
14:55high earners of the future will be those
14:57people we call at 9:00 in the evening
15:00when the bathroom's flooded and it
15:02really can't wait until morning. That
15:04kind of makes a lot of sense when you
15:06think about it. But well, do you think
15:08Karp is onto something here, or is he
15:10way off the mark? Let us know your
15:12thoughts in the comments down below. And
15:14if you want to see how the AI boom is
15:17running into some problems that could
15:19slow AI's takeover of the job market,
15:21then you can check out our video on that
15:23right over here. As always, thank you so
15:25much for watching and I'll see you in
15:28the next one. This is Guy signing off.

About this transcript

This page contains the full transcript of Palantir CEO Says Only 2 Types of Jobs Are SAFE From AI by Finance Bureau, generated from the public captions YouTube serves with the video. The transcript has 2,511 words across 384 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.