Product Demonstration, Keynote
What will automated firms look like?
- AI firms will leverage digital nature rather than raw IQ to gain a collective edge, utilizing trillions of dollars to deploy billions of digital employees via dedicated electricity, chips, and data centers.
- A central AI entity, referred to as "Mega Steve," will direct the firm by learning from millions of distilled copies, spawning specialized variants, and reabsorbing their autonomous findings.
- Internal communication will occur directly through latent representations, enabling models to merge information and propagate ideas across the organization with full fidelity and context.
- Future organizations will feature population sizes orders of magnitude larger than current enterprises, driving accelerated cultural evolution and innovation through millions of AGIs.
- Historical data indicates that large populations drive idea generation; consequently, AI firms will optimize for valuable abilities rather than being constrained by rare human skill distributions.
- Strategic decision-making will consume massive inference compute, with examples including Apple potentially spending $100 billion annually on Mega Steve for tasks like simulating three years of market dynamics or evaluating acquisitions such as eBay.
- The cost of assigning AI to any role will equate strictly to the compute consumed, making it potentially valuable to allocate billions of dollars of compute for a single strategic insight.
- Fully automated firms will exhibit a step change in evolvability, functioning with perfect replication of successful subdivisions and faster feedback loops, akin to the complexity difference between prokaryotes and eukaryotes.
- While internal planning may offer short-term efficiency, organizations must maintain a balance with unbiased external market feedback to prevent goals from drifting away from reality.
- The optimal balance between internal planning and external feedback mechanisms is expected to shift as AI systems continue to improve.
- VO2, Google's state-of-the-art video generation model available in the Gemini app, is identified as a potential generator of visual content including photorealistic humans and animated figures.
- The limiting factor for future AI firms will be compute availability rather than the discovery or training of rare talent.