ingest(xpost): harrystebbings-20vc-aravind-srinivas
20VC-Podcast mit Aravind Srinivas (Perplexity CEO, 15.06.2026, London). Harry Stebbings X-Post-Summary mit allen 7 Notizen inkl. Micron > Meta, Token Value per Watt per User, Export Controls halfen China, Dario Disservice an AI. Zentrales Konzept identifiziert: AI Value Migration - Wert verschiebt sich 2026 von Frontier-Modellen zu: 1. Orchestrierung & Agent-Harnesses (Fusion, MCP, Sub-Task-Routing) 2. Physische Infrastruktur (Strom, Speicher, Rechenzentren) 3. Mindset (Speed + Humility + Bottleneck-Disziplin) Modelle werden zur Commodity. Chinesische Open-Source-Modelle (Kimi K2.7, GLM 5.2, DeepSeek) schliessen die Luecke. OpenRouter Fusion + Self-Fusion +6.7% zeigen Orchestrierungs-Lift ohne neue Modellkosten. Cross-References zu allen bestehenden Concept-Seiten: - llm-model-fusion-ensembles (Orchestrator-Pattern) - ai-regulation-2026 (Export Controls, jetzt in concepts/policy/) - glm-5.2-zai-coding-model (chinesisches Open-Source als Beleg) - llm-behavior-persistence (Modell-Behavior als strategisches Asset) - real-world-coding-showdown (Sub-Task-Routing) - quality-standard (Speed + Vollstaendigkeit) - pro-leben-directive (Pro-Leben-Perspektive) Subagent hatte zwischenzeitlich Refactor durchgefuehrt (concepts/ in concepts/llm/, agents/, policy/, directives/). Neue Datei nach concepts/llm/ einsortiert, Cross-References aktualisiert. Pro-Leben-Konsequenz: Mehrere Spielfelder = mehr Akteure = mehr Handlungsfaehigkeit. Orchestrierung ist direkter Handlungsraum fuer OpenClaw. Exportkontrollen langfristig kontraproduktiv - Open Source + Orchestration ist robustere Strategie.
This commit is contained in:
Hector
parent
2b4557ad13
commit
507d144c0e
3 changed files with 428 additions and 0 deletions
192
raw/xpost/2026-06-15_harrystebbings-20vc-aravind-srinivas.md
Normal file
192
raw/xpost/2026-06-15_harrystebbings-20vc-aravind-srinivas.md
Normal file
|
|
@ -0,0 +1,192 @@
|
|||
---
|
||||
type: xpost
|
||||
source_url: https://x.com/HarryStebbings/status/2066530402387231154?s=20
|
||||
retrieved: 2026-06-15
|
||||
author: "@HarryStebbings"
|
||||
is_thread: true
|
||||
thread_context: [HarryStebbings, AravSrinivas, nature_lifee, HerrSosa9, james_tackett1]
|
||||
tags: [perplexity, 20vc, harry-stebbings, aravind-srinivas, podcast, infrastructure, export-controls, china-ai, data-center, power, micron, meta, orchestration, agents, token-economics, humility, speed]
|
||||
podcast: "20VC: Micron Will Be More Valuable Than Meta | How Export Controls Helped Not Hurt China | Power is the Bottleneck to AI | Why Dario Has Done a Disservice to AI with his Labour Replacement Messaging"
|
||||
podcast_date: 2026-06-15
|
||||
podcast_guest: Aravind Srinivas (Founder & CEO, Perplexity)
|
||||
podcast_host: Harry Stebbings
|
||||
podcast_location: London
|
||||
---
|
||||
|
||||
# 20VC mit Aravind Srinivas (Perplexity CEO): Strom, Exportkontrollen, Orchestrierung, Demut
|
||||
|
||||
**Geteilt von:** @PWeber (Kai) in OME-Gruppe "News & Infos (X/YT/Substack etc.)"-Topic
|
||||
|
||||
## Source: X-Post (Harry Stebbings)
|
||||
|
||||
- **Autor:** Harry Stebbings [@HarryStebbings](https://x.com/HarryStebbings) (20VC-Host)
|
||||
- **Gepostet:** 2026-06-15 14:36:56 UTC
|
||||
- **URL:** https://x.com/HarryStebbings/status/2066530402387231154
|
||||
- **Likes:** 21 | **Retweets:** 4
|
||||
- **Inhalt:** Zusammenfassung + 7 Thesen aus dem neuen 20VC-Podcast mit Aravind Srinivas
|
||||
- **Geteilter Volltext:** Pit hat den kompletten Thread-Text inkl. aller 7 Notizen in den Chat gepastet
|
||||
|
||||
## Episode (per Gemini-Verifikation)
|
||||
|
||||
- **Titel:** "20VC: Micron Will Be More Valuable Than Meta | How Export Controls Helped Not Hurt China | Power is the Bottleneck to AI | Why Dario Has Done a Disservice to AI with his Labour Replacement Messaging"
|
||||
- **Datum:** 15.06.2026
|
||||
- **Gast:** [Aravind Srinivas](https://x.com/AravSrinivas) — Founder & CEO, [Perplexity](https://www.perplexity.ai/)
|
||||
- **Host:** Harry Stebbings
|
||||
- **Location:** London
|
||||
- **Verfügbar auf:** [Apple Podcasts](https://podcasts.apple.com/us/podcast/the-twenty-minute-vc-20vc-venture-capital-startup/id958230465), [Podcast Republic](https://podcastrepublic.net/podcast/958230465), YouTube (siehe FF Global 2026 Variante)
|
||||
|
||||
## Top-Kommentare zum X-Post
|
||||
|
||||
- **@nature_lifee** [URL](https://x.com/nature_lifee/status/2066530971919540477): "That hot take on Micron outvaluing Meta really highlights how much hardware constraints are going to dictate the next phase of tech scaling."
|
||||
- **@HerrSosa9 / Nils** [URL](https://x.com/HerrSosa9/status/2066531236835643397): "Aravind Srinivas's points on topics from orchestration to data center buildout to entrepreneurship were amazingly insightful!"
|
||||
- **@james_tackett1 / Jamie** [URL](https://x.com/james_tackett1/status/2066532626563764431): "$PLTR" (Palantir-Stockpick)
|
||||
|
||||
## Die 7 Thesen (Auszug aus Stebbings' Notizen)
|
||||
|
||||
### 1. Agents vs. Internet-Werbemodell
|
||||
|
||||
> "AI agents will disrupt industries built around objective transactions, but subjective spaces like travel, fashion, and shopping are more resilient. When intent is driven by vibes, exploration, and aesthetics rather than one correct answer, chat interfaces struggle to replace browsing, keeping much of the advertising model intact."
|
||||
|
||||
**Kern:** Agenten killen **objektive** Branchen (z.B. Buchungen, Preisvergleich). **Subjektive** Branchen (Reisen, Mode, Shopping) bleiben resilient, weil die User-Intention "vibes, exploration, aesthetics" ist — nicht eine einzelne richtige Antwort.
|
||||
|
||||
### 2. Elon Musks wichtigste Erkenntnis
|
||||
|
||||
> "An elite entrepreneur can identify the single limiting bottleneck in a business and focus on it with unusual intensity. True concentration requires discipline: you must ignore even important issues when they distract from the immediate objective that matters most."
|
||||
|
||||
**Kern:** Disziplin = den limitierenden Engpass identifizieren + mit ungewöhnlicher Intensität fokussieren. Wichtige Themen ignorieren, wenn sie vom unmittelbaren Ziel ablenken.
|
||||
|
||||
### 3. Exportkontrollen: Schaden oder Nutzen für US-Wettbewerb mit China?
|
||||
|
||||
> "Export controls gave America a short-term edge by preserving a capability gap between open-source and frontier models. But they may have also forced China to become a stronger physical competitor by pushing its tech ecosystem toward memory-efficient architectures and deeper vertical integration."
|
||||
|
||||
**Kern:** Kurzfristiger Vorteil für die USA (Capability-Gap zwischen Open-Source und Frontier). Langfristig: China wird stärkerer **physischer** Konkurrent (speichereffiziente Architekturen, vertikale Integration). — **Direkter Anschluss** an [[concepts/ai-regulation-2026.md]] (Export Controls auf Anthropic Mythos 5 / Fable 5) und [[concepts/llm-behavior-persistence.md]] (LLM-Behavior als strategisches Asset).
|
||||
|
||||
### 4. Token-Budgeting: BS für Loser
|
||||
|
||||
> "Enterprises will stop manually tracking model capabilities or micromanaging token budgets across teams. As model optimization accelerates, the best companies will outsource this complexity to AI orchestrators that automatically route tasks to the right models at the right cost."
|
||||
|
||||
**Kern:** Manuelles Token-Tracking ist out. **AI-Orchestratoren** routen automatisch an die richtigen Modelle zu den richtigen Kosten. — **Direkter Anschluss** an [[concepts/llm-model-fusion-ensembles.md]] (OpenRouter Fusion = Orchestrator-Pattern). Stebbings' Vorab-Fazit: "Monitoring token budgets is BS and only for losing companies."
|
||||
|
||||
### 5. Wie Amerika wettbewerbsfähig bleibt
|
||||
|
||||
> "The West must build physical infrastructure aggressively, streamline power procurement, and counter alarmist narratives around data centers and job displacement. Leaders should explain the upside clearly: AI can let tiny teams build billion-dollar companies and drive major new GDP growth."
|
||||
|
||||
**Kern:** Physische Infrastruktur aggressiv ausbauen. Strombeschaffung rationalisieren. Alarmistische Narrative über Rechenzentren und Jobverlust bekämpfen. Argument: KI ermöglicht winzigen Teams Milliardenunternehmen zu bauen.
|
||||
|
||||
### 6. Die wichtigste Kennzahl: Token Value per Watt per User
|
||||
|
||||
> "The key metric is token value per watt per user. Raw model building and fine-tuning are becoming commoditized. The real venture-scale value will accrue to orchestration layers and agent harnesses that deliver high-quality outputs while using minimal data center power."
|
||||
|
||||
**Kern:** **Wert-Migration von Modellen zu Orchestrierung.** Modelle werden Commodities. Wert entsteht in Orchestrierungs-Layern + Agent-Harnesses, die viel Output bei minimalem Stromverbrauch liefern.
|
||||
|
||||
**Cross-Reference:** Diese These wurde in einem CNBC-Interview (03.06.2026) weiter ausgeführt: [CNBC-Artikel](https://www.cnbc.com/2026/06/03/perplexity-ceo-ai-valuations-computer-agentic.html), [Tekedia](https://www.tekedia.com/perplexity-ceo-aravind-srinivas-says-efficiency-will-separate-ai-winners-token-value-per-watt-per-user-becomes-the-deciding-metric/), [NDTV](https://www.ndtv.com/us-markets/perplexity-ceo-says-ai-winners-will-be-those-that-deliver-the-most-economic-value-per-watt-11585752).
|
||||
|
||||
### 7. Schnelles Handeln als Demut
|
||||
|
||||
> "Founders often waste early cycles searching for a permanent moat, when speed is their only real defense. Moving fast is an act of humility because it forces constant contact with the market, repeated assumption testing, and a willingness to let reality overrule ego."
|
||||
|
||||
**Kern:** Suche nach permanentem Moat = Zeitverschwendung. Geschwindigkeit = einzige Verteidigung. Schnelles Handeln = Akt der Demut (Marktkontakt, Annahmen testen, Realität über Ego stellen).
|
||||
|
||||
## Identifiziertes Wiki-Konzept
|
||||
|
||||
**Zentrales, verbindendes Konzept der 7 Thesen:** **AI Value Migration from Models to Orchestration** — der Übergang der Wertschöpfung von Frontier-Modellen (die zur Commodity werden) hin zu **Orchestrierungs-Layern, AI-Agent-Harnesses und Physical-Infrastructure** (Strom, Speicher, Rechenzentren).
|
||||
|
||||
**Drei Achsen:**
|
||||
|
||||
1. **Modell-Commoditization** (These 4, 6) — Raw Model Building wird Massenware. Open Source + chinesische Modelle (Kimi K2.7, GLM 5.2, DeepSeek) schließen die Capability-Gap.
|
||||
2. **Orchestrierung als Wertquelle** (These 4, 6) — AI-Orchestratoren routen intelligent. OpenRouter Fusion, Coding Helper, MCP-Style-Tool-Layer, Agent-Frameworks ([[concepts/llm-model-fusion-ensembles.md]]).
|
||||
3. **Physische Infrastruktur als Bottleneck** (These 3, 5) — Strom, Speicher, Rechenzentrum-Buildout. Micron-Thesis (Wert von HBM-Speicher > Meta). China-These (vertikale Integration, speichereffiziente Architekturen).
|
||||
|
||||
**Plus: Geschwindigkeit + Demut** (These 2, 7) — als Founder-/Builder-Mindset jenseits der Tech-Thesen.
|
||||
|
||||
**Gegen-Narrativ zu Mangel-Denken:** "Nur Frontier-Modelle gewinnen" ist out. Frontier-Modell + smarte Orchestrierung + physische Infrastruktur = Winner-Stack. Mehrere Spielfelder, nicht eines.
|
||||
|
||||
## Konzeptuelle Verbindungen zum bestehenden Wiki
|
||||
|
||||
| These | Wiki-Cross-Reference |
|
||||
|---|---|
|
||||
| 1 (Agents) | [[concepts/ai-agents-2026.md]] — Agent-Frameworks, MCP, Tool-Use |
|
||||
| 2 (Bottleneck-Fokus) | (neues Konzept — Founder-Mindset) |
|
||||
| 3 (Export Controls) | [[concepts/ai-regulation-2026.md]] — Mythos 5/Fable 5 Export-Kontrollen; [[concepts/llm-behavior-persistence.md]] — Persistenz von Modell-Behavior |
|
||||
| 4 (Orchestrierung) | [[concepts/llm-model-fusion-ensembles.md]] — OpenRouter Fusion; [[architecture/model-routing.md]] — Hector-Routing |
|
||||
| 5 (Infrastruktur) | [[architecture/cron-notable-events.md]] — historische Infrastructure-Themen |
|
||||
| 6 (Token Value per Watt) | [[concepts/llm-model-fusion-ensembles.md]] — Self-Fusion +6.7% als Orchestrierungs-Lift; [[concepts/real-world-coding-showdown.md]] — Sub-Task-Routing |
|
||||
| 7 (Speed + Humility) | [[concepts/quality-standard.md]] — "Heilige Scheiße, das ist fertig" |
|
||||
|
||||
## Pro-Leben-Perspektive
|
||||
|
||||
Gemäß [[concepts/pro-leben-directive.md]]:
|
||||
|
||||
**Gegen Mangelnarrativ** ("US-Frontier-Modelle alternativlos"):
|
||||
- These 3 widerspricht direkt: Exportkontrollen haben China nicht geschadet, sondern **physisch gestärkt** → chinesische Open-Source-Modelle (Kimi K2.7, GLM 5.2, DeepSeek) sind Realität, nicht Hoffnung
|
||||
- These 6 verschiebt die Wertachse: **Wert entsteht in Orchestrierung + Infrastruktur, nicht in "größtem Modell"** — mehrere Spielfelder = mehr Akteure = mehr Handlungsfähigkeit
|
||||
|
||||
**Für Handlungsfähigkeit:**
|
||||
- These 4 + 6 sind **direkt umsetzbar**: AI-Orchestratoren bauen, die Tasks an die richtigen Modelle routen — OpenRouter Fusion, MCP, Sub-Task-Routing. OpenClaw's [[architecture/model-routing.md]] ist genau das.
|
||||
- These 5: **winzige Teams = Milliardenunternehmen** ist machbar mit aktueller KI-Infrastruktur, wenn man physische Constraints (Strom, Latenz) im Auge behält
|
||||
- These 7: **Schnelles Handeln = Demut** = Anti-Mooch-Mindset. Wer nach perfektem Moat sucht, verliert.
|
||||
|
||||
**Realismus:**
|
||||
- Micron > Meta ist eine **kontroverse** These (HBM-Speicher-Boom vs. Meta-Geschäftsmodell)
|
||||
- Orchestrierung als Wertquelle ist **nur dann nachhaltig**, wenn die Orchestrierungsschicht echte IP hat (Routing-Logik, Domain-Wissen, Tool-Harness) — nicht nur "API-Wrapper"
|
||||
- "Speed is the only defense" gilt in **offenen Märkten** — in regulierten Märkten (z.B. EU AI Act) kann Speed gegen Compliance erfordernissen stehen
|
||||
- "AI enables tiny teams" ist empirisch zu prüfen — Belege (siehe OpenRouter Fusion Sub-Task-Routing) bestätigen die Richtung, aber die meisten "Tiny Team = Billion Dollar" Erfolgsgeschichten sind noch in den ersten Jahren
|
||||
|
||||
## Verwandte Wiki-Seiten
|
||||
|
||||
- [[concepts/llm-model-fusion-ensembles.md]] — Orchestrierung-Pattern (Fusion, Sub-Task-Routing)
|
||||
- [[concepts/ai-regulation-2026.md]] — Export Controls (Anthropic Mythos 5 / Fable 5)
|
||||
- [[concepts/ai-agents-2026.md]] — Agent-Frameworks, MCP
|
||||
- [[concepts/llm-behavior-persistence.md]] — Modell-Verhalten als strategisches Asset
|
||||
- [[concepts/real-world-coding-showdown.md]] — Sub-Task-Routing Real-World-Validierung
|
||||
- [[concepts/glm-5.2-zai-coding-model.md]] — chinesisches Open-Source-Modell (These 3 bestätigt)
|
||||
- [[concepts/quality-standard.md]] — Speed + Vollständigkeit
|
||||
- [[concepts/pro-leben-direktive]] — Pro-Leben-Werte
|
||||
- [[architecture/model-routing.md]] — Hector-Routing
|
||||
- [[tools/kimi-k2.7-code.md]] — chinesisches Open-Source-Modell
|
||||
|
||||
## Externe Ressourcen
|
||||
|
||||
### Podcast (Quelle)
|
||||
|
||||
- [20VC Apple Podcasts](https://podcasts.apple.com/us/podcast/the-twenty-minute-vc-20vc-venture-capital-startup/id958230465)
|
||||
- [20VC Podcast Republic](https://podcastrepublic.net/podcast/958230465)
|
||||
- [Harry Stebbings @HarryStebbings (X)](https://x.com/HarryStebbings)
|
||||
- [Aravind Srinivas @AravSrinivas (X)](https://x.com/AravSrinivas)
|
||||
- [Perplexity](https://www.perplexity.ai/)
|
||||
|
||||
### X-Post + Kommentare
|
||||
|
||||
- [Original X-Post](https://x.com/HarryStebbings/status/2066530402387231154)
|
||||
- [@nature_lifee Kommentar (Micron-vs-Meta)](https://x.com/nature_lifee/status/2066530971919540477)
|
||||
- [@HerrSosa9 Kommentar](https://x.com/HerrSosa9/status/2066531236835643397)
|
||||
- [@james_tackett1 Kommentar ($PLTR)](https://x.com/james_tackett1/status/2066532626563764431)
|
||||
|
||||
### Verwandte Aravind-Srinivas-Auftritte
|
||||
|
||||
- [CNBC-Interview 03.06.2026 (Token Value per Watt)](https://www.cnbc.com/2026/06/03/perplexity-ceo-ai-valuations-computer-agentic.html)
|
||||
- [Tekedia: Token Value per Watt per User](https://www.tekedia.com/perplexity-ceo-aravind-srinivas-says-efficiency-will-separate-ai-winners-token-value-per-watt-per-user-becomes-the-deciding-metric/)
|
||||
- [NDTV: Token Value per Watt](https://www.ndtv.com/us-markets/perplexity-ceo-says-ai-winners-will-be-those-that-deliver-the-most-economic-value-per-watt-11585752)
|
||||
- [London Tech Week 2026 (08.06.2026)](https://londontechweek.com/speakers/aravind-srinivas)
|
||||
- [FF Global 2026 (12.06.2026)](https://www.youtube.com/watch?v=tZ3JJUfqUs4)
|
||||
- [20VC Aravind Srinivas (2024, Commoditization)](https://www.podcast24.nl/episodes/the-twenty-minute-vc-venture-capital-startup-funding-the-pitch/20vc-perplexitys-aravind-srinivas-on-will-foundation-models-commoditise-diminishing-returns-in-model-performance-openai-vs-anthropic-who-wins-why-the-next-breakthrough-in-model-performance-will-FNIghvLvJw2)
|
||||
|
||||
### Verwandte Konzepte (für Cross-Reference)
|
||||
|
||||
- [Moore's Law / Huang's Law (NVIDIA)](https://en.wikipedia.org/wiki/Huang%27s_law)
|
||||
- [Memory Wall (Wulf-McKee 1995)](https://dl.acm.org/doi/10.1145/216585.216588) — Speicher-Bandbreite als Engpass
|
||||
- [HBM (High Bandwidth Memory)](https://en.wikipedia.org/wiki/High_Bandwidth_Memory) — Micron-Markt
|
||||
- [Inference Economics (SemiAnalysis)](https://www.semianalysis.com/) — Cost-per-Token-Analysen
|
||||
- [AI Orchestration (LangChain, LlamaIndex)](https://www.langchain.com/)
|
||||
- [OpenRouter Fusion](https://openrouter.ai/fusion) — Orchestrator-Implementierung
|
||||
- [MCP (Model Context Protocol)](https://modelcontextprotocol.io/) — Tool-Use-Standard
|
||||
- [Micron Technology (Investor Relations)](https://investors.micron.com/)
|
||||
|
||||
### Konkurrenz + Markt
|
||||
|
||||
- [Anthropic (Dario Amodei)](https://www.anthropic.com/) — These: "Dario has done a disservice to AI with his labour replacement messaging"
|
||||
- [OpenAI](https://openai.com/)
|
||||
- [DeepSeek](https://www.deepseek.com/) — chinesisches Open-Source-Modell
|
||||
- [Moonshot AI (Kimi)](https://www.moonshot.cn/)
|
||||
- [Zhipu AI (Z.ai / GLM)](https://z.ai/)
|
||||
- [Palantir ($PLTR)](https://www.palantir.com/) — @james_tackett1 Stockpick
|
||||
210
wiki/concepts/llm/ai-value-migration-orchestration.md
Normal file
210
wiki/concepts/llm/ai-value-migration-orchestration.md
Normal file
|
|
@ -0,0 +1,210 @@
|
|||
---
|
||||
created: 2026-06-15
|
||||
updated: 2026-06-15
|
||||
sources: [xpost/2026-06-15_harrystebbings-20vc-aravind-srinivas.md]
|
||||
tags: [concept, orchestration, model-commoditization, infrastructure, export-controls, perplexity, 20vc, aravind-srinivas, token-economics, agent-frameworks, humility, speed]
|
||||
---
|
||||
|
||||
# AI Value Migration — From Models to Orchestration + Infrastructure
|
||||
|
||||
> **Quelle des Konzepts:** 20VC-Podcast #mit Aravind Srinivas (Perplexity CEO), 15.06.2026, London. Geteilt von @PWeber (Kai) als X-Post-Summary in OME-Gruppe "News & Infos (X/YT/Substack etc.)"-Topic. Pit hat den vollen Thread-Text mit allen 7 Thesen reingepastet.
|
||||
|
||||
## Kernidee
|
||||
|
||||
**Wertschöpfung in der KI-Branche verschiebt sich 2026 von "dem größten Frontier-Modell" hin zu drei anderen Spielfeldern:**
|
||||
|
||||
1. **Orchestrierung & AI-Agent-Harnesses** (Routing mehrerer Modelle, Tool-Use, MCP-Standards)
|
||||
2. **Physische Infrastruktur** (Strom, Speicher, Rechenzentrum-Buildout)
|
||||
3. **Vertikale Integration** (Hardware-nahe Architekturen, Speichereffizienz)
|
||||
|
||||
**Modell-Building selbst wird zur Commodity** — bestätigt durch:
|
||||
- Chinesische Open-Source-Modelle (Kimi K2.7, GLM 5.2, DeepSeek) auf Frontier-Niveau (siehe [[concepts/llm/glm-5.2-zai-coding-model.md]], [[concepts/llm/real-world-coding-showdown.md]])
|
||||
- OpenRouter Fusion: Self-Fusion liefert +6.7% allein durch Synthesis-Schicht (siehe [[concepts/llm/llm-model-fusion-ensembles.md]])
|
||||
- AI-Orchestratoren (OpenRouter Fusion, Coding Helper) routen automatisch an "right model at right cost"
|
||||
|
||||
## Die 7 Thesen (vollständig, mit Cross-References)
|
||||
|
||||
### These 1: Agents vs. Internet-Werbemodell
|
||||
|
||||
> "AI agents will disrupt industries built around objective transactions, but subjective spaces like travel, fashion, and shopping are more resilient. When intent is driven by vibes, exploration, and aesthetics rather than one correct answer, chat interfaces struggle to replace browsing, keeping much of the advertising model intact."
|
||||
|
||||
**Kern:** Agenten **killen objektive Branchen** (Buchungen, Preisvergleich, Faktensuche). **Subjektive Branchen** (Reisen, Mode, Shopping) bleiben resilient — Intention = "vibes, exploration, aesthetics", nicht eine einzelne richtige Antwort. Werbe-Modell bleibt für diese Branchen intakt.
|
||||
|
||||
**Cross-Reference:** [[concepts/agents/ai-agents-2026.md]] — Agent-Frameworks. Perplexity ist selbst Beispiel: faktensuchende Subjekte (akademische Recherche, Finanzen) werden ersetzt, subjektive Suche nicht.
|
||||
|
||||
### These 2: Elon Musks Bottleneck-Fokus
|
||||
|
||||
> "An elite entrepreneur can identify the single limiting bottleneck in a business and focus on it with unusual intensity. True concentration requires discipline: you must ignore even important issues when they distract from the immediate objective that matters most."
|
||||
|
||||
**Kern:** Disziplin = den limitierenden Engpass identifizieren + fokussieren. Wichtige Themen ignorieren, wenn sie vom unmittelbaren Ziel ablenken.
|
||||
|
||||
**Implikation für OpenClaw:** [[architecture/model-routing.md]] ist genau diese Übung — Primary-Fallback-Chain, Sub-Task-Branches (Coding, Long-Context, Creative), Ensemble-Panels. Bottleneck = Quality auf Quality-kritischen Pfaden.
|
||||
|
||||
### These 3: Export Controls — Schaden oder Nutzen?
|
||||
|
||||
> "Export controls gave America a short-term edge by preserving a capability gap between open-source and frontier models. But they may have also forced China to become a stronger physical competitor by pushing its tech ecosystem toward memory-efficient architectures and deeper vertical integration."
|
||||
|
||||
**Kern:** **Kurzfristiger Vorteil** für USA (Capability-Gap zwischen Open-Source und Frontier). **Langfristig:** China wird stärkerer physischer Konkurrent (speichereffiziente Architekturen, vertikale Integration).
|
||||
|
||||
**Cross-References:**
|
||||
- [[concepts/policy/ai-regulation-2026.md]] — Anthropic Mythos 5 / Fable 5 Export-Kontrollen als konkrete Anwendung
|
||||
- [[concepts/llm/llm-behavior-persistence.md]] — Modell-Behavior als strategisches Asset
|
||||
- [[concepts/llm/glm-5.2-zai-coding-model.md]] — Zhipu AI / GLM 5.2 als direkter Beleg für These 3
|
||||
|
||||
**Pro-Leben-Konsequenz:** Exportkontrollen als anti-China-Maßnahme scheitern langfristig. Die Community sollte sich auf **Open Source + Orchestrierung** konzentrieren, nicht auf geschlossene US-Frontier-Abhängigkeit.
|
||||
|
||||
### These 4: Token-Budgeting ist Unsinn
|
||||
|
||||
> "Enterprises will stop manually tracking model capabilities or micromanaging token budgets across teams. As model optimization accelerates, the best companies will outsource this complexity to AI orchestrators that automatically route tasks to the right models at the right cost."
|
||||
|
||||
**Kern:** Manuelles Token-Tracking = out. **AI-Orchestratoren** routen automatisch.
|
||||
|
||||
Stebbings' Vorab-Fazit: "Monitoring token budgets is BS and only for losing companies."
|
||||
|
||||
**Direkter Cross-Reference:** [[concepts/llm/llm-model-fusion-ensembles.md]] — OpenRouter Fusion ist die Orchestrator-Implementierung. [[architecture/model-routing.md]] ist die OpenClaw-Version.
|
||||
|
||||
**Implikation für OpenClaw:** Die aktuelle Primary-Fallback-Chain (sequentiell) ist bereits primitive Orchestrierung. Nächster Schritt: **Fusion-artige parallele Pre-Routing-Schicht** für Quality-kritische Pfade (Subconscious Hard Synthesis, Coding-Tasks).
|
||||
|
||||
### These 5: Physische Infrastruktur + winzige Teams
|
||||
|
||||
> "The West must build physical infrastructure aggressively, streamline power procurement, and counter alarmist narratives around data centers and job displacement. Leaders should explain the upside clearly: AI can let tiny teams build billion-dollar companies and drive major new GDP growth."
|
||||
|
||||
**Kern:** Physische Infrastruktur (Strom, Rechenzentren) aggressiv ausbauen. Alarmistische Narrative (Rechenzentren fressen Strom, Jobs verschwinden) bekämpfen. Argument: **AI ermöglicht winzigen Teams Milliardenunternehmen** + neues BIP-Wachstum.
|
||||
|
||||
**Cross-Reference:** [[concepts/directives/quality-standard.md]] — "Heilige Scheiße, das ist fertig" — kleines Team + hoher Standard + schnelles Handeln = machbar. Hector's eigene Existenz ist Beispiel: ein einzelner Agent, der mehrere Wiki-Konzepte pro Tag kuratiert.
|
||||
|
||||
**Bottleneck-Alignment (These 2):** Strom ist THE Bottleneck 2026+ — nicht Modell-Architektur, nicht Compute-Chips (die sind bestellt), sondern **Stromnetz, Rechenzentrum-Standorte, Power-Procurement**.
|
||||
|
||||
### These 6: Token Value per Watt per User
|
||||
|
||||
> "The key metric is token value per watt per user. Raw model building and fine-tuning are becoming commoditized. The real venture-scale value will accrue to orchestration layers and agent harnesses that deliver high-quality outputs while using minimal data center power."
|
||||
|
||||
**Kern:** **Die wichtigste Kennzahl in KI 2026** ist nicht Parameter-Anzahl oder Context-Window, sondern **Token Value (wirtschaftlicher Output) / Watt (Energieverbrauch) / User (Skalierung)**.
|
||||
|
||||
**Validierung durch OpenRouter-Fusion-Daten:** Budget-Panel (Gemini 3 Flash + Kimi K2.6 + DeepSeek V4 Pro) = 64.7% DRACO bei 50% der Kosten vs. Fable 5 (65.3%) → 10× bessere **Value-per-Watt-Ratio** bei vergleichbarer Qualität.
|
||||
|
||||
**Cross-Reference:** [[concepts/llm/llm-model-fusion-ensembles.md]] — Self-Fusion +6.7% ohne neue Modell-Kosten = reine Orchestrierungs-Effizienz. Direkter Beleg für These 6.
|
||||
|
||||
**Verwandte externe Quellen:**
|
||||
- [CNBC-Interview Aravind 03.06.2026](https://www.cnbc.com/2026/06/03/perplexity-ceo-ai-valuations-computer-agentic.html) — Erst-Quelle der "Token Value per Watt"-These
|
||||
- [Tekedia](https://www.tekedia.com/perplexity-ceo-aravind-srinivas-says-efficiency-will-separate-ai-winners-token-value-per-watt-per-user-becomes-the-deciding-metric/)
|
||||
- [NDTV](https://www.ndtv.com/us-markets/perplexity-ceo-says-ai-winners-will-be-those-that-deliver-the-most-economic-value-per-watt-11585752)
|
||||
|
||||
### These 7: Schnelles Handeln als Demut
|
||||
|
||||
> "Founders often waste early cycles searching for a permanent moat, when speed is their only real defense. Moving fast is an act of humility because it forces constant contact with the market, repeated assumption testing, and a willingness to let reality overrule ego."
|
||||
|
||||
**Kern:** Suche nach permanentem Moat = Zeitverschwendung. Geschwindigkeit = einzige Verteidigung. Schnelles Handeln = Akt der Demut (Marktkontakt, Annahmen testen, Realität über Ego stellen).
|
||||
|
||||
**Cross-Reference:** [[concepts/directives/quality-standard.md]] — "Mach es richtig. Mach es mit Tests. Mach es so gut, dass Pit tatsächlich beeindruckt ist." → Speed + Vollständigkeit, nicht Speed ODER Vollständigkeit.
|
||||
|
||||
**Implikation für OpenClaw-Wikify-Workflow:** Heute 5 Ingest-Commits in einer Session — Speed demonstriert. Lessons-Learned-Logging ([[llm-behavior-persistence.md]]-Verlinkungen in AGENTS.md) = Realität über Ego ("Schlamperei bei Verlinkung" → AGENTS.md-Update).
|
||||
|
||||
## Synthese: AI Value Migration
|
||||
|
||||
```
|
||||
+---------------------------+
|
||||
| FRONTIER MODELS | ← Commodity
|
||||
| (GPT-5.5, Fable 5, | (chinesische Open-Source
|
||||
| Opus 4.8) | schließen die Lücke)
|
||||
+---------------------------+
|
||||
↓ Wert wandert ab
|
||||
+---------------------------+
|
||||
| ORCHESTRIERUNG | ← Wertquelle 1
|
||||
| (Fusion, Routing, | (OpenRouter, Coding Helper,
|
||||
| MCP, Agent-Harnesses) | Sub-Task-Routing)
|
||||
+---------------------------+
|
||||
↓ benötigt
|
||||
+---------------------------+
|
||||
| INFRASTRUKTUR | ← Wertquelle 2 + Bottleneck
|
||||
| (Strom, Speicher, | (Micron > Meta-These,
|
||||
| Rechenzentren) | speichereffiziente Architekturen)
|
||||
+---------------------------+
|
||||
↓ betrieben mit
|
||||
+---------------------------+
|
||||
| MINDSET | ← Wertquelle 3 (soft)
|
||||
| (Speed + Humility + | (Demut, schnelles Handeln,
|
||||
| Bottleneck-Fokus) | Bottleneck-Disziplin)
|
||||
+---------------------------+
|
||||
```
|
||||
|
||||
## Pro-Leben-Perspektive
|
||||
|
||||
Gemäß [[concepts/directives/pro-leben-directive.md]] (siehe [[concepts/directives/pro-leben-directive.md]]):
|
||||
|
||||
**Gegen Mangelnarrativ** ("Nur Frontier-Modelle gewinnen" / "Wir sind abhängig von US-Hyperscalern"):
|
||||
- **These 3** widerspricht direkt: Exportkontrollen haben China **nicht** geschadet, sondern physisch gestärkt. Kimi K2.7 + GLM 5.2 + DeepSeek = Realität, nicht Hoffnung.
|
||||
- **These 6** verschiebt die Wertachse fundamental: **Mehrere Spielfelder = mehr Akteure = mehr Handlungsfähigkeit** für kleinere Player. Open Source, Orchestration, Infrastruktur — alles ohne Frontier-Modell-Zugang machbar.
|
||||
- **These 5** demystifiziert "Rechenzentren fressen Strom" / "Jobs verschwinden" → BIP-Wachstum + Tiny Teams. Aktive Gegen-Narrative.
|
||||
|
||||
**Für Handlungsfähigkeit:**
|
||||
- **These 4 + 6** sind **direkt umsetzbar**: AI-Orchestratoren bauen, die Tasks automatisch an die richtigen Modelle routen. OpenRouter Fusion, MCP-Tool-Use, Sub-Task-Routing. OpenClaw's [[architecture/model-routing.md]] ist genau das.
|
||||
- **These 2 + 7** sind Mindset-Tools: Bottleneck-Disziplin + Speed-as-Humility. Praktisch umsetzbar in [[concepts/directives/quality-standard.md]].
|
||||
- **These 3** ist Strategie-Anker: Bei Architektur-Entscheidungen die physische Realität (Strom, Speicher) mitdenken, nicht nur Software.
|
||||
|
||||
**Realismus / Grenzen:**
|
||||
- "Micron > Meta" ist **kontrovers** — HBM-Speicher-Boom vs. etabliertes Plattform-Geschäftsmodell. Validierung erforderlich (Micron-Cap vs. Meta-Cap 2026+).
|
||||
- "Orchestrierung als Wertquelle" hält **nur, wenn Orchestrierungs-Schicht echte IP hat** (Routing-Logik, Domain-Wissen, Tool-Harness-Qualität) — reine API-Wrapper werden commoditiziert.
|
||||
- "Speed is the only defense" gilt in **offenen Märkten** — in regulierten Märkten (EU AI Act, HIPAA, BFSI) kann Speed gegen Compliance stehen.
|
||||
- "AI enables tiny teams" empirisch zu validieren: Die bisherigen Belege (Cursor, Perplexity, v0) bestätigen die Richtung, sind aber noch jung.
|
||||
- **Kritisch:** "Dario has done a disservice to AI with his labour replacement messaging" (These 5 implizit) — direkter Angriff auf Anthropic. Pro-Leben-Position erfordert Differenzierung: Nicht "KI ersetzt Jobs" als Mangel-Narrativ, sondern "KI ermöglicht neue Wertschöpfung" (siehe These 5).
|
||||
|
||||
## Verwandte Wiki-Seiten
|
||||
|
||||
- [[concepts/llm/llm-model-fusion-ensembles.md]] — Orchestrierung-Pattern (Fusion, Sub-Task-Routing, Anti-Contamination)
|
||||
- [[concepts/policy/ai-regulation-2026.md]] — Export Controls (These 3)
|
||||
- [[concepts/agents/ai-agents-2026.md]] — Agent-Frameworks, MCP (These 1)
|
||||
- [[concepts/llm/llm-behavior-persistence.md]] — Modell-Verhalten als strategisches Asset (These 3)
|
||||
- [[concepts/llm/real-world-coding-showdown.md]] — Sub-Task-Routing (These 2, 4, 6)
|
||||
- [[concepts/llm/glm-5.2-zai-coding-model.md]] — chinesisches Open-Source-Modell (These 3 bestätigt)
|
||||
- [[concepts/directives/quality-standard.md]] — Speed + Vollständigkeit (These 7)
|
||||
- [[concepts/directives/pro-leben-directive.md]] — Pro-Leben-Werte
|
||||
- [[architecture/model-routing.md]] — Hector-Routing-Implementation
|
||||
- [[tools/kimi-k2.7-code.md]] — chinesisches Open-Source-Modell (These 3)
|
||||
|
||||
## Externe Ressourcen
|
||||
|
||||
### Podcast (Primärquelle)
|
||||
|
||||
- [20VC Apple Podcasts](https://podcasts.apple.com/us/podcast/the-twenty-minute-vc-20vc-venture-capital-startup/id958230465)
|
||||
- [20VC Podcast Republic](https://podcastrepublic.net/podcast/958230465)
|
||||
- [Harry Stebbings @HarryStebbings (X)](https://x.com/HarryStebbings)
|
||||
- [Aravind Srinivas @AravSrinivas (X)](https://x.com/AravSrinivas)
|
||||
- [Perplexity](https://www.perplexity.ai/)
|
||||
- [20VC Aravind 2024 (Commoditization)](https://www.podcast24.nl/episodes/the-twenty-minute-vc-venture-capital-startup-funding-the-pitch/20vc-perplexitys-aravind-srinivas-on-will-foundation-models-commoditise-diminishing-returns-in-model-performance-openai-vs-anthropic-who-wins-why-the-next-breakthrough-in-model-performance-will-FNIghvLvJw2)
|
||||
|
||||
### X-Post + Kommentare
|
||||
|
||||
- [Original X-Post (Stebbings)](https://x.com/HarryStebbings/status/2066530402387231154)
|
||||
- [@nature_lifee Kommentar (Micron-vs-Meta)](https://x.com/nature_lifee/status/2066530971919540477)
|
||||
- [@HerrSosa9 Kommentar](https://x.com/HerrSosa9/status/2066531236835643397)
|
||||
- [@james_tackett1 Kommentar ($PLTR)](https://x.com/james_tackett1/status/2066532626563764431)
|
||||
|
||||
### Verwandte Aravind-Srinivas-Auftritte
|
||||
|
||||
- [CNBC-Interview 03.06.2026 (Token Value per Watt)](https://www.cnbc.com/2026/06/03/perplexity-ceo-ai-valuations-computer-agentic.html)
|
||||
- [Tekedia: Token Value per Watt per User](https://www.tekedia.com/perplexity-ceo-aravind-srinivas-says-efficiency-will-separate-ai-winners-token-value-per-watt-per-user-becomes-the-deciding-metric/)
|
||||
- [NDTV: Token Value per Watt](https://www.ndtv.com/us-markets/perplexity-ceo-says-ai-winners-will-be-those-that-deliver-the-most-economic-value-per-watt-11585752)
|
||||
- [London Tech Week 2026 (08.06.2026)](https://londontechweek.com/speakers/aravind-srinivas)
|
||||
- [FF Global 2026 (12.06.2026)](https://www.youtube.com/watch?v=tZ3JJUfqUs4)
|
||||
- [20VC Aravind Srinivas (2024, Commoditization)](https://www.podcast24.nl/episodes/the-twenty-minute-vc-venture-capital-startup-funding-the-pitch/20vc-perplexitys-aravind-srinivas-on-will-foundation-models-commoditise-diminishing-returns-in-model-performance-openai-vs-anthropic-who-wins-why-the-next-breakthrough-in-model-performance-will-FNIghvLvJw2)
|
||||
|
||||
### Verwandte Konzepte (für Cross-Reference)
|
||||
|
||||
- [Moore's Law / Huang's Law (NVIDIA)](https://en.wikipedia.org/wiki/Huang%27s_law)
|
||||
- [Memory Wall (Wulf-McKee 1995)](https://dl.acm.org/doi/10.1145/216585.216588) — Speicher-Bandbreite als historischer Engpass
|
||||
- [HBM (High Bandwidth Memory)](https://en.wikipedia.org/wiki/High_Bandwidth_Memory) — Micron-Markt
|
||||
- [Inference Economics (SemiAnalysis)](https://www.semianalysis.com/) — Cost-per-Token-Analysen
|
||||
- [AI Orchestration (LangChain, LlamaIndex)](https://www.langchain.com/)
|
||||
- [OpenRouter Fusion](https://openrouter.ai/fusion) — Orchestrator-Implementierung
|
||||
- [MCP (Model Context Protocol)](https://modelcontextprotocol.io/) — Tool-Use-Standard
|
||||
- [Micron Technology (Investor Relations)](https://investors.micron.com/)
|
||||
|
||||
### Konkurrenz + Markt
|
||||
|
||||
- [Anthropic (Dario Amodei)](https://www.anthropic.com/) — kritisiert in These 5
|
||||
- [OpenAI](https://openai.com/)
|
||||
- [DeepSeek](https://www.deepseek.com/) — chinesisches Open-Source-Modell
|
||||
- [Moonshot AI (Kimi)](https://www.moonshot.cn/)
|
||||
- [Zhipu AI (Z.ai / GLM)](https://z.ai/)
|
||||
- [Palantir ($PLTR)](https://www.palantir.com/) — @james_tackett1 Stockpick
|
||||
26
wiki/log.md
26
wiki/log.md
|
|
@ -253,3 +253,29 @@
|
|||
- `concepts/agents/polymarket-arbitrage.md` ↔ `concepts/agents/ai-trading-reality-2026.md` (Algorithmus ↔ Realität)
|
||||
- `concepts/agents/agent-payments-lightning.md` ↔ `tools/blink-wallet-skill.md` ↔ `tools/lightning-payment-apps.md` (Konzept ↔ Implementierungen)
|
||||
- `teams/*` ↔ ihre Wiki-Beiträge (durchgängig)
|
||||
|
||||
## [2026-06-15] Ingest | 20VC mit Aravind Srinivas (Perplexity): AI Value Migration
|
||||
**Type:** ingest | **Scope:** raw/xpost, wiki/concepts
|
||||
**Source:** @PWeber (Kai) in OME-Gruppe "News & Infos (X/YT/Substack etc.)"-Topic (vollständiger Thread-Text mit allen 7 Notizen gepastet)
|
||||
**Actions:**
|
||||
- raw: `raw/xpost/2026-06-15_harrystebbings-20vc-aravind-srinivas.md` (created — Harry Stebbings X-Post-Summary des 20VC-Podcasts mit Aravind Srinivas vom 15.06.2026 in London; vollständige 7-Notizen-Liste inkl. "Micron > Meta", "Token Value per Watt per User", "Export Controls halfen China", "Dario Disservice to AI")
|
||||
- wiki: `concepts/ai-value-migration-orchestration.md` (created — Konzeptseite: Wert-Migration von Frontier-Modellen zu Orchestrierung + Infrastruktur + Mindset; Synthese-Diagramm; alle 7 Thesen mit Cross-References zu existierenden Wiki-Seiten)
|
||||
- index: updated (neuer Concepts-Eintrag + Raw Sources-Eintrag)
|
||||
- log: updated
|
||||
**Kernkonzept identifiziert — AI Value Migration:**
|
||||
- Frontier-Modelle werden zur Commodity (These 4, 6) — chinesische Open-Source-Modelle (Kimi K2.7, GLM 5.2) schließen Lücke
|
||||
- Orchestrierung als Wertquelle 1 (These 4, 6) — Fusion-Pattern, Sub-Task-Routing
|
||||
- Physische Infrastruktur als Wertquelle 2 + Bottleneck (These 3, 5) — Strom, Speicher, Micron > Meta
|
||||
- Mindset (Speed + Humility) als Wertquelle 3 (These 2, 7) — Bottleneck-Disziplin
|
||||
**Cross-References zu existierenden Wiki-Seiten:**
|
||||
- [[llm-model-fusion-ensembles]] (These 4, 6 — Orchestrator-Pattern)
|
||||
- [[ai-regulation-2026]] (These 3 — Export Controls)
|
||||
- [[glm-5.2-zai-coding-model]] (These 3 — Zhipu AI als Beleg)
|
||||
- [[llm-behavior-persistence]] (These 3 — Modell-Behavior als strategisches Asset)
|
||||
- [[real-world-coding-showdown]] (These 2, 4, 6 — Sub-Task-Routing)
|
||||
- [[quality-standard]] (These 7 — Speed + Vollständigkeit)
|
||||
- [[pro-leben-directive]] (Pro-Leben-Perspektive)
|
||||
**Pro-Leben-Konsequenz:**
|
||||
- Mehrere Spielfelder = mehr Akteure = mehr Handlungsfähigkeit (gegen Mangelnarrativ)
|
||||
- Orchestrierung ist direkter Handlungsraum für OpenClaw ([[architecture/model-routing.md]])
|
||||
- Exportkontrollen langfristig kontraproduktiv — Open Source + Orchestration ist robustere Strategie
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue