← Home
Thompson argues that the companies winning the AI race face a paradox: the more they invest in AI infrastructure, the more they commit to a technology whose value proposition is not yet clear. The winner of the AI race may be cursed by overinvestment.
This summary was generated from show notes and public descriptions, not from a full transcript review. Details may contain inaccuracies.
Canon
•
Thompson identifies the AI capability treadmill: GPT-3 was amazing in 2020. GPT-4 was amazing in 2023. GPT-5 will be amazing briefly. Each generation raises the baseline, requiring ever-more-expensive training runs to maintain the lead.
Highlights
•
The winner's curse in AI: the company that spends the most on AI infrastructure wins the race but may have overpaid for a prize of uncertain value
Thompson applies auction theory to the AI race: when multiple bidders compete for an asset of uncertain value, the winner typically overpays because winning requires bidding above what the asset is rationally worth. AI infrastructure spending may follow this pattern.Was this useful?