TThe Diary of a CEO
← All episodes
Eric Schmidt14 November 2024

Ex Google CEO: AI Is Creating Deadly Viruses! If We See This, We Must Turn Off AI! They Leaked Our Secrets At Google!

9Frameworks
14Insights

Frameworks in this episode

Mindset5 steps

Check the Assertion

Distinguish being marketed to from being given an argument — and verify before you repeat

Read →
Leadership4 steps

Divas Not Knaves

Ally yourself with the brilliant, difficult person who wants the product perfect — and screen out the self-server

Read →
Leadership4 steps

Good Revenue vs Bad Revenue

Split every quality gain 50/50 between monetisation and product — bad revenue is not sustainable

Read →
Leadership5 steps

Hire for Risk, Not Resume

In a startup, prioritise intelligence and quickness over experience — and never hire someone you'd later lay off

Read →
Strategy4 steps

The 70/20/10 Rule

70% core business, 20% adjacent, 10% new ideas — the allocation that paid for every Google failure

Read →
Strategy5 steps

The AI Tripwires: When to Pull the Plug

Three specific, observable conditions at which humans should assert control over an AI system

Read →
Leadership4 steps

The Pirate Building

You cannot harvest and hunt in the same room — put the disruptors in a different building with different rules

Read →
Strategy5 steps

The Scale Test

Only back ideas that can go from zero to infinity — then get to 100 million users and monetisation follows

Read →
Strategy4 steps

Write Down Five Years, Then Run One

Predict the world in five years in writing, then set hard one-year goals against it and measure them

Read →

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Myth Buster· 1

Myth Buster1:42:00

'I want people in an office — but the data doesn't support us'

Schmidt argues hard for offices, especially for people in their twenties: he learned by hanging out at the water cooler, in hallways and in meetings, and says that knowledge was central to his promotions. Then he does something unusual and concedes the evidence cuts against him. The data indicates productivity is actually slightly higher when you allow work from home, at least in the industries that have been studied. He does not like it, but he acknowledges the science is there.

  • If you're in your twenties, being in an office is how you get promoted.
  • Remote-work objections about commuting and family are real issues.
  • Studied industries show productivity slightly higher with work-from-home flexibility.
  • Most companies have landed on hybrid — two or three days rather than full rollback.

“the problem with our joint view is it's not supported by the data”

Eric Schmidt · 1:42:30

“I don't happen to like it but I want to acknowledge the science is there”

Eric Schmidt · 1:43:30
#remote-work#office#productivity#evidence

Hot Take· 2

Hot Take14:00

We're running an experiment on a billion people with no control group

Schmidt argues adults are largely safe from AI reprogramming — your values and the way you get up in the morning are already set — but children are not. The core developmental question of the AI revolution is what happens to a child's identity when their best friend from birth is a computer. Nobody knows, because it has never been done. He pairs it with the documented epidemic of harm to teenage girls, who get hit by social media at 11 and 12 and drive record emergency-room and self-harm figures.

  • Adult values are unlikely to be rewritten by AI; children's can be.
  • Nobody knows what a computer best friend does to a child's identity.
  • Girls get hit at 11 and 12, before they can handle the emotional load.
  • Society is starting to respond — French schools banning classroom phones.

“you're running an experiment on a billion people without a control”

Eric Schmidt · 14:30
#children#ai-risk#social-media-harm#development
Hot Take1:09:30

AI clones of your podcast won't replace you — they'll amplify you

Bartlett worries billions of AI-generated podcasts will erode his moat. Schmidt says the evidence is against it, and tells him to move from scarcity thinking to abundance: have the fake podcasters criticise him and his guests, annotate and amplify his interviews, and reach audiences that love them more than him — but all derived from him. He points to the same prediction failing at Google, where people expected celebrity to dissolve into a long tail of micro-markets and instead networks accentuated the best people and made them global.

  • Scale computing generates abundance, and abundance allows new strategies.
  • Networks accentuated the best people rather than flattening them into a long tail.
  • You went from local to national to global personality — and the globe is big.
  • Used well, AI makes you more famous, not less.

“what's really going on is you're moving from scarcity to ubiquity you're moving from scarc to abundance”

Eric Schmidt · 1:09:30

“if you do it well by using these AI Technologies you will become more famous not less famous”

Eric Schmidt · 1:12:00
#media#ai-content#abundance#creator-economy

Story· 3

Story38:00

The pirate flag: why you can't harvest and hunt in the same building

Asked whether a team can innovate while running the day job, Schmidt says there are almost no examples of doing both simultaneously in the same building. His case is Steve Jobs putting the small Macintosh team in a little building beside the main one in Cupertino with a pirate flag on top. Culturally it was bad — it created resentment in the big building — but it was absolutely right for Apple's revenue and path, because the Mac established the user interface that eventually made the iPhone possible.

  • Almost no examples exist of harvesting and hunting in the same building.
  • The pirate team must be exempt from the incumbent's rules to move fast.
  • Resentment from the main building is a real, acceptable cost.
  • The Mac's UI lineage is what made the iPhone possible.

“you can't get people to play two roles the incentives are different if you're going to be a pirate and a disruptor you don't have…”

Eric Schmidt · 39:00
#innovation#org-design#apple#disruption
Story43:30

Intel sold ARM — the simplification rule that cost them mobile forever

Schmidt uses Intel to attack the business-school rule that you should simplify product lines and cut what doesn't fit. Intel sold off the ARM RISC architecture because it was incompatible with their main chip architecture. That architecture turned out to be exactly what mobile phones needed — low memory, small batteries, heat constraints — so Intel was never a player in mobile. Today Nvidia's B200 pairs its GPUs with an ARM CPU, not Intel's. The lesson: simple rules require a written model of the next five years.

  • Intel sold ARM ~15 years ago as an architecture-fit simplification.
  • ARM was what mobile needed: low memory, small battery, low heat.
  • Nvidia's B200 uses an ARM CPU, not an Intel one.
  • They failed to see that battery power would matter as much as computing power.

“the core decision which was to simplify simplify to the wrong outcome”

Eric Schmidt · 44:30
#strategy#intel#arm#forecasting
Story48:30

ChatGPT was an afterthought — even OpenAI didn't know what they had

Schmidt's account of how OpenAI got ahead: the Transformer was invented at Google, but an OpenAI team figured out RLHF — using humans at the end to run A/B judgments so the system learns recursively from human training. That was the real breakthrough, and Schmidt says none of them expected it, himself included. His joke to his OpenAI friends is that they turned it on one Thursday night and realised how good it was. They were working on GPT-4 at the time; ChatGPT was an afterthought that became a success disaster.

  • The T in GPT — Transformer — was invented at Google.
  • RLHF, humans A/B-judging outputs so the model learns recursively, was the breakthrough.
  • Even brilliant founders don't necessarily understand how powerful their work is.
  • Google's defensive answer: it was busy running eight to ten billion user clusters of activity.

“they didn't really understand how good it was they just turned it on and all of a sudden they had this huge success disaster”

Eric Schmidt · 50:30
#openai#rlhf#google#ai-history

tactic· 3

tactic07:30

If you can't tell true from false, keep your mouth shut

Schmidt defines critical thinking operationally: distinguish being marketed to — which he equates with being lied to — from being handed the argument to evaluate yourself. The practical habit he built at Google is checking any plausible-sounding claim rather than believing it because friends do. He uses the widely repeated but false statistic that only 10% of Americans have passports as the model case, and frames repeating unverified claims as a personal responsibility failure.

  • Being marketed to is a form of being lied to.
  • Plausibility is why false claims spread — so check the plausible ones.
  • You are responsible for the truth of anything you repeat.
  • Falsifiability is what makes science trustworthy: people constantly test it.

“if somebody says something plausible just check it”

Eric Schmidt · 09:00

“you have a responsibility before you repeat something to make sure what you're repeating is true and if you can't distinguish between true and false…”

Eric Schmidt · 09:00
#critical-thinking#misinformation#epistemics#media-literacy
tactic16:00

Good revenue vs bad revenue: the arbitrary 50/50 rule at Google

Every time Google improved search quality, Schmidt faced the same call — convert the gain into more ads, or into a better product. He arbitrarily set the split at 50/50 because he judged both to matter, with the founders' backing, and his summary is that Google became more moral and also made more money. The alternative model, surfacing lies and deception because they draw people in, fails on principle and on sustainability: Gresham's law applies to speech, and bad speech drives out good.

  • A fixed 50/50 split removes the argument from every individual launch.
  • Bad revenue is not just wrong, it is unsustainable.
  • Google's real design win: bad results exist, but not on the first page.
  • Gresham's law applied to speech — bad speech drives out good speech.

“I arbitrarily decided that we would take 50% to one 50% to the other because I thought they were both important”

Eric Schmidt · 16:30

“so Google became more moral and also made more money”

Eric Schmidt · 16:30
#monetisation#product-decisions#platform-ethics#google
tactic57:30

The 70/20/10 rule and the ten-person team that made tens of billions

Schmidt lays out the allocation rule Larry and Sergey devised: 70% on the core business of search and ads, 20% on adjacent businesses like cloud, 10% on genuinely new ideas. The 10% bucket produced Google X, whose first product was Google Brain — one of the first machine-learning architectures, predating DeepMind. A team of ten or fifteen people generated tens of billions of dollars in extra profit over a decade, which is what makes fast failure affordable everywhere else in the portfolio.

  • 70% core, 20% adjacent, 10% new — Larry and Sergey's split.
  • Google X's first product was Google Brain, which preceded DeepMind.
  • Ten to fifteen people generated tens of billions over a decade.
  • Silicon Valley's real advantage: you can spend years on a bad idea, get cancelled, and get hired again.

“at Google we had this 72010 rule that Larry and Sergey came up with 70% of the Core Business 20% on adjacent business and 10%…”

Eric Schmidt · 57:30

“my joke is the best CFO is one who's just gone bankrupt because the one thing that CFO is not going to let happen is…”

Eric Schmidt · 58:30
#resource-allocation#innovation#google#fast-failure

insight· 2

insight10:30

Why platforms converge on outrage: the objective function explains everything

Schmidt explains social media harm mechanically rather than morally. Algorithms literally maximise a mathematical goal, and in this case that goal is attention. To maximise revenue you maximise attention, and the cheapest way to maximise attention is to maximise outrage. He describes TikTok as a bandit algorithm — keep serving what you signal you want, occasionally sample the adjacent area — which reliably produces rabbit holes of confirmatory content, and notes that an unhappy user's whole online environment becomes unhappy people.

  • Algorithms mathematically maximise a trained objective; here it is attention.
  • Revenue → attention → outrage is the chain, not a moral failing of any one CEO.
  • TikTok works like a Las Vegas one-armed bandit: exploit, occasionally explore adjacent.
  • Algorithms won't offer a struggling user positive alternatives unless forced to.

“the easiest way to maximize attention is to maximize outrage”

Eric Schmidt · 12:00
#algorithms#social-media#attention-economy#objective-function
insight1:36:00

The three tripwires where humans should pull the plug on AI

Schmidt rejects the idea that AGI arrives on a particular day and instead names concrete intervention points. First, recursive self-improvement where you no longer know what the system is learning. Second, a system producing a new model faster than the previous one could be checked. Third, an agent announcing it will invent a private language only other agents understand. To the objection that we can't unplug them, his answer is literal: there's a power plug and a circuit breaker, go turn it off.

  • AGI as a single-day event is unlikely; capability arrives in waves per field.
  • Tripwire 1: recursive self-improvement you can no longer inspect.
  • Tripwire 2: new models produced faster than the last one was verified.
  • Tripwire 3: agents inventing a language only agents understand.

“at some point if you don't know what it's learning you should unplug it”

Eric Schmidt · 1:36:00

“one of the agents says I have a better idea I'm going to communicate in my own language that I'm going to invent that only…”

Eric Schmidt · 1:37:30
#ai-safety#agents#agi#oversight

contrarian· 3

contrarian46:00

Crypto is not a platform wave — AI is

Schmidt separates genuine platform waves from important-but-specialised markets. The PC and the internet were broad enough to create a whole new ocean; he argues crypto is not, because crypto is not transformative to daily life for everyone and people are not running around all day using tokens instead of currency. It is important and interesting, but a specialised market rather than a horizontal one. The arrival of what he calls alien intelligence in the form of a savant you use is the transformative thing, because it touches everything.

  • Platform waves create a whole new ocean; crypto did not.
  • Crypto is important and interesting but a specialised, not horizontal, market.
  • AI touches producers, executives, narratives, markets — everything.
  • The practical question for any business: how will you apply AI to accelerate what you're doing?

“crypto is not transformative to daily life for everyone people are not running around all day using crypto tokens rather than currency”

Eric Schmidt · 46:30
#crypto#platforms#ai#technology-waves
contrarian1:01:00

Why lay people off — just don't hire them in the first place

Against the then-standard practice of culling the bottom 5% every six to nine months, Google took the opposite position: don't lay people off, don't hire them to begin with. Schmidt's objection to routine culls is twofold — you are assuming the bottom 5% has been correctly identified, and even the lowest performers hold knowledge and value the corporation can use. In his entire decade as CEO the only layoff was roughly 200 people in sales after the 2000 crash, and he remembers it as extremely painful.

  • Google's stance: don't lay off, just don't hire — much, much easier.
  • Routine 5% culls assume the ranking is accurate; it usually isn't.
  • Even the lowest performers hold knowledge and value.
  • One layoff in a decade: ~200 sales roles after the 2000 crash.

“we had a position of why lay people off just don't hire them in the first place it's much much easier”

Eric Schmidt · 1:01:00
#hiring#layoffs#management#google-culture
contrarian1:37:30

His actual fear isn't rogue AI — it's that we adopt it too slowly

Asked for his biggest fear about AI, Schmidt says it differs from what you'd expect: his real fear is that we won't adopt it fast enough to solve the problems that affect everybody. People want safety, healthcare and good schools. He asks why we don't yet have an AI teacher working alongside human teachers in the child's own language and culture, or a doctor's assistant that knows every possible best treatment given inventory, country and insurance. He calls these relatively achievable, and says education and healthcare alone would establish a global level playing field of knowledge and opportunity.

  • His real fear is under-adoption, not runaway AI.
  • An AI teacher in the child's own language and culture is achievable now.
  • A doctor's assistant that knows every best treatment given real constraints is achievable now.
  • Education plus healthcare globally would establish a level playing field of opportunity.

“my actual fear is we're not going to adopt it fast enough to solve the problems that affect everybody”

Eric Schmidt · 1:37:30
#ai-optimism#education#healthcare#adoption