Interview, Fireside Chat
Tomasz Tunguz: How I Raised $230M; ChatGPT vs. Google; How LLMs Work; Trump vs DeSantis | E1004
- The fundraising market is expected to remain challenging for the next couple of months, with a standard timeline of 12 months for founders to understand a space, including six to nine months of research, while the venture capital environment over the next decade will differ significantly from the previous ten, necessitating consistent investment to provide founder comfort.
- The speaker anticipates that by the end of 2023, the US will reach its lowest point due to Federal Reserve rate overcorrection and a faster-than-expected decrease in money production (M1 and M2), driven by a meaningful recession risk compounded by significant geopolitical tensions in Taiwan.
- Economic outlooks project the US GDP currently growing at roughly 2% annually, with AI potentially increasing this by 2.5% and doubling GDP growth over a 40 to 60-year period to replicate post-war prosperity, though entitlement spending is forecast to consume 95% of tax receipts within 10 years, requiring government reform.
- Regulatory costs are expected to benefit incumbents, citing a rise from a $25 million public offering cost at $25 million revenue in the mid-90s to $15 million at $100 million revenue today, with the regulatory system evolving incrementally where regulatory friction acts as a price of progress.
- Diversification benefits are projected to be 82 to 84% achieved with a portfolio of 23 companies, while private market valuations have remained elevated relative to down public markets, requiring LPs more time to understand their portfolios.
- Public market multiples for top quartile companies have fallen from 40x during quantitative easing to approximately 6x currently, creating a significant disconnect between public and private market valuations.
- The tech sector faces a massive opportunity for LLMs to destabilize the B2B internet, creating a niche for B2C-focused funds, while the search industry faces an innovator's dilemma where Google must choose between self-disruption or being displaced by competitors.
- Future AI development will likely see 75 to 80% of new code generation driven by artificial intelligence within 10 years, with 70 to 80% of all code generation being AI-driven as most code consists of slight modifications to standard existing code.
- Generative model query costs are currently 100 to 10,000 times higher than search queries, but a geometric curve of improvement is anticipated alongside the emergence of "emergent behavior" in LLMs that creates compounding benefits.
- Market architecture is expected to bifurcate with data remaining in customer accounts while the application plane runs on software company servers, leading to a future consumer model mediated by a single interface across different AI models and an enterprise model focused on platform providers like Stripe and Twilio.
- Customer behavior is projected to shift from requesting bundled end-to-end solutions in early markets to preferring best-of-breed layers as sophistication increases, while startups can still win against incumbents through superior execution despite the "David versus Goliath" dynamic.
- Content ownership and data attribution landscapes will necessitate TBD revenue share arrangements as AI operators seek viable business models for content producers, and the relationship between governments and people is expected to change meaningfully over the next decade similar to trends in France.
- The Republican Party is expected to remain the party of business and capitalism, tasked with guiding reforms for the impending entitlement crisis, while the cost of regulation and the risk of conflict in Taiwan add to the odds of a US recession.