George Sivulka, Co-Founder & CEO @Hebbia: The Future of Foundation Models | E1250

Hebbia CEO George Sivulka argues RAG is fundamentally broken and that scaling compute at inference — not training — is the next frontier for enterprise AI.

• Hebbia raised a $130M Series B; investors include a16z, Peter Thiel, Index, and GV.
• Sivulka argues RAG fails because ~90% of enterprise queries require reasoning about documents, not retrieval from them.
• Hebbia pioneered “scaling at inference”: running hundreds or thousands of parallel model calls per query instead of relying on a larger model or longer context.
• Sivulka believes xAI is the most undervalued AI company and may overtake OpenAI and Anthropic in value within 12–24 months.
• Foundation model layer will commoditize; Nvidia’s dominance will erode as inference shifts to AMD and custom ASICs away from CUDA-dependent training workflows.
• Excel reached 90% penetration in finance in 18–24 months (1985–86) — Sivulka uses this as evidence that finance, often cited as slow-moving, adopts fast when real value is demonstrated.
• Sivulka dismisses Klarna’s AI headcount-reduction claims as marketing, arguing loud displacement narratives signal internal anxiety rather than real change.

2025-01-22 · Watch on YouTube