# schmidt-tangen-superintelligence-in-good-company-2026-10-21

## Veille

Episode of the video podcast **In Good Company** by **Norges Bank Investment Management** (Norwegian sovereign wealth fund), published on YouTube on **October 21, 2026**, **43 minutes**, in English. **Nicolai Tangen**, the fund's CEO, interviews **Eric Schmidt**, former CEO of Google (2001-2011).

## Titre Article

Eric Schmidt: Are We Ready for Superintelligence? | Podcast | In Good Company

## Date

2026-10-21

## URL

https://www.youtube.com/watch?v=U5YtR3OG5Uo

## Keywords

superintelligence, AGI, scientific discovery, recursive self-improvement, reinforcement learning, AI agents, Claude Tag, organizational flattening, coordination tax, layoffs, time compression, Anthropic's growth, Europe, regulation, European foundation model, Mistral AI, ASML, labor law, modèles chinois, distillation, open weights, guardrails, international treaties, mini Chernobyl, Claude Mythos, disinformation, democracy, child protection, founder myth, science funding

## Authors

Eric Schmidt (invité), ancien PDG de Google ; Nicolai Tangen (hôte), directeur général de Norges Bank Investment Management.

## Ton

Profile: an executive-to-executive interview, conducted by an institutional investor for an audience of investors and business leaders. Short, open-ended questions; the host rarely pushes back and punctuates with approval.

Register: assertive, first-person, rich with anecdotes (the employee "Vinnie" directed by his boss's AI avatar, the friend convinced that data centers are killing cattle, the VAX 11/780 at Berkeley). Schmidt claims his own bluntness (*"I'm going to sound like an incredibly obnoxious American and I am"*) and owns blunt statements about headcount and about Europe.

Evidence: few sourced figures. The estimates (5-10 years for AGI, six months of Chinese lag, 20% of staff laid off at a friend's company, two thousand superfluous positions out of ten thousand) are presented as personal judgments. Several facts are mentioned without being named: "10 major math problems" solved in a month, a bot incident a week earlier.

Positioning: Schmidt discloses his own interests himself — an investor in Mistral, a builder of data centers, head of four companies. He describes himself as optimistic, except about democracy, and closes on his own shortcomings.

## Pense-betes

- **Source**: In Good Company podcast (NBIM), October 21, 2026, 43 min; Eric Schmidt interviewed by Nicolai Tangen.
- **Five waves**: mainframe, PC, Internet, social networks, AI; each larger, AI *"touches everything."* Driver: fundamental discoveries + compute at scale (2nm chips, memory, data centers).
- **Not a bubble**: Anthropic has more than doubled Google's revenue growth record.
- **Model cycle**: 6 months to build, 2 to test, obsolete in 5; costs hundreds of millions of dollars. *"The next cycle will drive people insane because it'll be days."*
- **Current frontier**: AI writes all the code (*"My job is dead"*) and knows everything known; discovery remains out of reach. Two paths: recursive self-improvement / RL "monkey at the keyboard," or transferring patterns across domains. AGI in **5-10 years**.
- **Fast gains** in *scale-free* domains: software, cyber, math; slow where data is scarce (biology).
- **Agents**: input → reasoning + activities → synthesis; every business chain can be described as concatenated agents, then agents + robots.
- **Organization**: Claude Tag in Slack, the CEO querying everyone's tags; elimination of *"glue people"*; need for inventors, salespeople, executives, and government relations.
- **Europe**: regulates before innovating; underfunded universities; the 21-employee threshold; expensive energy, training to be done at partner countries; ASML as proof of capability; captive European market for Mistral. Context in [[sfeir-rapport-kill-switch-souverainete-2026-07-24]].
- **China**: distillation, smuggled chips, **six months** behind, free models; GLM (Zhipu) with capabilities close to Mythos, cf. [[fasano-fleischer-anthropic-glm-5-3-advanced-cyber-capabilities-2026-09-29]].
- **Unresolved risk**: removal of guardrails from an open-weight model.
- ⚠️ **Scope**: personal estimates, major facts left unnamed (math problems, bot incident).
- **Advice for young people**: build an agentic system and a foundation model for one's field.

## RésuméDe400mots

In the Norges Bank Investment Management podcast In Good Company, published on October 21, 2026, Nicolai Tangen interviews Eric Schmidt, former CEO of Google. Schmidt has lived through five technology waves: mainframe, PC, Internet, social networks, AI. Each is larger than the one before; AI is larger still because it touches everything. He dismisses the idea of a bubble: Anthropic has more than doubled the revenue growth record set by Google. Time is compressing: a model is built in six months, tested in two, and obsolete within five.

AI can already write all the code and gather all known knowledge, but it cannot yet discover: a model trained on 1902 mathematics does not rediscover general relativity. Two paths exist — random exploration through reinforcement learning, and transferring patterns from one domain to another; Schmidt places AGI five to ten years out. Progress is fastest in software, cyber, and mathematics, domains without data constraints, and slowest in biology.

Agents chain together query, reasoning, actions, and synthesis; nearly every professional activity can be decomposed into concatenated agents. With tools like Claude Tag, the company becomes flatter, faster, more intense. Schmidt doubts that large companies, especially European ones, will change their culture. He estimates that a company of 10,000 people has two thousand superfluous positions, to be transitioned legally and humanely.

Europe regulates before it innovates, lacks capital, well-funded universities, and cheap energy, and has no foundation model of its own, which leaves its history and culture to American models or modèles chinois. It can dominate its own market, as Mistral does, in which Schmidt is an investor, and ASML shows it can integrate a supply chain. He proposes applying only health and safety rules to companies with fewer than 21 employees.

China is said to be six months behind, thanks to distillation and smuggled chips; its free models will win over users in poorer countries. The suspension of Mythos upset his European friends. The worst-case scenario is a ruinous race between China and the West with a loss of control; treaties will likely follow an accident, a "mini Chernobyl." The removal of guardrails from open-weight models remains unresolved.

He describes himself as pessimistic about democracy, exposed to online fake friends, and calls for protecting children. He now wants to fund science at scale.

## GrapheDeConnaissance

- Nicolai Tangen —dirige→ Norges Bank Investment Management (ORGANISATION, 0.98)
- Norges Bank Investment Management —publie→ podcast In Good Company (DOCUMENT, 0.97)
- Eric Schmidt —dirige→ Google (ORGANISATION, 0.95)
- Eric Schmidt —mesure→ croissance du chiffre d'affaires d'Anthropic plus de deux fois plus rapide que le record de Google (MESURE, 0.85)
- Eric Schmidt —affirme_que→ un modèle se construit en six mois, se teste en deux et est dépassé en cinq (AFFIRMATION, 0.88)
- Eric Schmidt —prédit→ l'AGI arrivera dans cinq à dix ans (AFFIRMATION, 0.9)
- Eric Schmidt —affirme_que→ l'IA ne sait pas encore découvrir : un modèle limité aux mathématiques de 1902 ne retrouve pas la relativité générale (AFFIRMATION, 0.9)
- auto-amélioration récursive —permet→ découverte scientifique (CONCEPT, 0.72)
- Eric Schmidt —affirme_que→ les progrès sont les plus rapides en logiciel, cyber et mathématiques, domaines sans contrainte de données (AFFIRMATION, 0.88)
- agents IA —s_applique_à→ workflows métier (CONCEPT, 0.85)
- Claude Tag —permet→ aplatissement organisationnel (CONCEPT, 0.8)
- Eric Schmidt —prédit→ une entreprise de 10 000 personnes compte environ deux mille postes dont elle n'a plus besoin (AFFIRMATION, 0.85)
- Eric Schmidt —affirme_que→ l'Europe régule avant d'innover et reste résistante au changement (AFFIRMATION, 0.92)
- Eric Schmidt —recommande→ n'appliquer aux entreprises de moins de 21 salariés que les règles de santé et de sécurité (AFFIRMATION, 0.88)
- Eric Schmidt —collabore_avec→ Mistral AI (ORGANISATION, 0.85)
- Eric Schmidt —affirme_que→ les modèles chinois ont six mois de retard, mais leur gratuité leur gagne la majorité des utilisateurs (AFFIRMATION, 0.88)
- modèles chinois —utilise→ Distillation (CONCEPT, 0.78)
- GLM —concurrence→ Claude Mythos (TECHNOLOGIE, 0.72)
- Eric Schmidt —prédit→ les traités sur l'IA suivront un accident de type « mini Chernobyl » (AFFIRMATION, 0.85)
- Eric Schmidt —affirme_que→ le retrait des garde-fous des modèles à poids ouverts est un problème non résolu (AFFIRMATION, 0.88)
- Eric Schmidt —affirme_que→ l'IA et les réseaux sociaux menacent la démocratie en fabriquant de faux amis en ligne (AFFIRMATION, 0.88)
- Eric Schmidt —recommande→ construire un système agentique et un modèle de fondation pour son domaine d'intérêt (AFFIRMATION, 0.85)

---
Canonical: https://www.thekb.eu/en/fiches/schmidt-tangen-superintelligence-in-good-company-2026-10-21/