AI summary by TwinMind
The AI opportunity: Sequoia Capital's AI Ascent 2024 opening remarks
Sequoia Capital
26m 56s
Opening and Market Opportunity Overview
- Pat Grady from Sequoia Capital introduces team members Sonia and Constantine as MCs
- AI has evolved from "magic box" to three distinct capabilities:
- Create (generative AI - images, text, video, audio)
- Reason (one-shot or multi-step agentic reasoning)
- Interact in humanlike capacity
- Cloud transition analogy: 2010 global software TAM $350B (cloud $6B), 2023 TAM $650B (cloud $400B)
- 40% CAGR over 15 years
- AI opportunity: replacing services with software vs cloud replacing software with software
- Starting point: tens of trillions vs hundreds of billions
- Hypothesis: "single greatest value creation opportunity known"
Technology Wave History and AI Timing
- Sequoia's historical perspective through technology waves:
- 1960s: Silicon-based transistors (Don Valentine at Fairchild)
- 1970s: Systems built on chips
- 1980s: Networks connecting PCs, packaged software
- 1990s: Internet - public-facing networks
- 2000s: Cloud - sophisticated internet applications
- 2010s: Mobile - apps in pockets
- AI concept dates to 1940s, neural nets emerged then
- Required ingredients now available:
- Cheap, plentiful compute
- Fast, efficient, reliable networks
- 7 of 8 billion people have supercomputers (smartphones)
- COVID forced everything online
- Data readily available
- Strong conviction: next 10-20 years will be "time of AI"
Current AI Market State and Examples
- Generative AI revenue: ~$3B aggregate (excluding FANG/cloud provider incremental revenue)
- SaaS market took nearly decade to reach $3B, generative AI achieved in first year
- Real-world AI implementations:
- Clara: 2/3 customer service inquiries automated, equivalent of 700 full-time agents
- Harvey: Legal services automation
- Software engineering: AI agents writing code
- AI avatars demo: Realistic virtual presence technology (HeyGen)
Investment and Funding Patterns
- Uneven funding distribution in AI "layer cake":
- Foundation models (bottom): Receiving most funding
- Developer tools/infrastructure (middle)
- Applications (top): Less funding than expected
- David's "$200B question": $50B spent on Nvidia GPUs vs $3B AI revenue
- "Math isn't mathing yet" - investment vastly exceeds returns
User Engagement and Retention Challenges
- AI apps show lower engagement than mobile peers:
- Lower daily/monthly active user ratios
- Poor one-month retention rates
- Gap between expectations and reality
- AI currently "once a week/month tinkery phenomenon"
- Recent model improvements: Sora, Claude 3, Grok
- Baseline intelligence rising should accelerate product-market fit
Mobile/iPhone Development Analogy
- iPhone 2007, App Store 2008
- Early apps: beer drinking, lightsaber, flip cup, flashlight
- Instagram and DoorDash didn't emerge until 2010
- Time needed to discover and harness new capabilities
- Same pattern expected for AI
Emerging AI Use Cases
- Customer support: First area hitting product-market fit
- AI friendship: Surprising application with strong user metrics
- Horizontal enterprise knowledge: Unlocking enterprise data
2024 Predictions
- Co-pilots to Agents: Transition from helper tools to autonomous systems
- Remove humans from loop entirely
- AI as coworker vs tool
- Focus areas: software engineering, customer service
- Enhanced Reasoning: Beyond statistical text patterns
- Inference time compute
- Game-play style value iteration
- Higher-level cognitive tasks (planning, reasoning)
- Enterprise Reliability: Evolution from consumer apps to high-stakes applications
- Hospitals, defense applications
- Tools for 5-9s reliability: RLHF, prompt chaining, vector databases
- Production Deployment: Prototypes moving to production
- Latency, cost, model ownership matter
- Compute shift from pre-training to inference
Long-term Future Vision: Productivity Revolution
- AI as productivity revolution (vs communication/transportation revolutions)
- Pattern: Human + tool → Human + machine assistant → Human + machine network
- Historical example: Sickle (10,000 years) → Mechanical reaper (1831) → Combined harvester (machine networks)
- AI agents working in complex networks
Software and Writing Evolution Examples
- Software development: Manual → Machine-assisted → Machine networks
- Writing: Human process → Human + assistant → Human + network of assistants
- Constantine's workflow: Multiple AI models (GPT-4, Mistral Large, Claude 3) working together
Economic Impact and Cost Reduction
- S&P 500 workers needed per $1M revenue declining rapidly
- "Faster and fewer" - not to do less, but to do more
- Deflationary areas: Computer software, accessories, televisions
- Inflationary areas needing AI help: Education, medical care, housing
Conceptual Computing Revolution
- Jensen Huang's prediction: Pixels will be generated, not rendered
- Shift from rote storage to conceptual representation
- Letter "A" example: From pixel matrix to multi-dimensional concept
- Computers understanding context, not just rendering
- Alignment with human thinking patterns
- Platonic forms concept (2,500 years ago)
Company-as-Neural-Network Vision
- Integration into specific processes and KPIs
- Customer support example: Fully connected optimization graph
- Layers of abstraction becoming interoperable
- Entire companies functioning like neural networks
- "One person company" enabling more problems to be tackled
Closing Vision
- AI positioned to drive down costs in crucial areas:
- Better education
- Healthier populations
- More productive populations
- Attendees will decide and build the future
- Goal: Abstract away complexity and mundane details for more powerful solutions
