Palantir CTO: AI Needs Diffusion, Not Better Models — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Shyam Sankar, Palantir's CTO, returned to Stratechery after a June 2023 interview to argue that labs and the wider market structure over-invest intellectually in better models and under-invest in diffusion—turning tokens into economic value faster.
- Sankar compares AI diffusion to electricity: power producers alone created little value until machine tools and factories were redesigned around electrons.
- Google's original “Attention Is All You Need” team, in Sankar's telling, pursued a 3% improvement in Google Translate, which he uses to illustrate how practical production goals can produce profound innovation.
- Sankar rejects outsourcing production as harmless comparative advantage, arguing that marginal reliance on WuXi can culminate in a transfer of knowledge and the loss of domestic capability.
- Palantir CTO Shyam Sankar argues that companies focused on the top line have greater success adopting AI; the interview also examines how firms can preserve alpha instead of the median.
- Anthropic's battle with the U.S. government is revisited by Sankar alongside his frustration that AI labs refuse to take a systems-engineering approach to safety.
- Mobilize, Sankar's book, carries the exhortation “To Reboot the American Industrial Base and Stop World War III,” while he argues for rebuilding the U.S. industrial base.
Why it matters: The economic center of gravity in Sankar's account is not model leaderboards but the companies and industrial systems that put AI to work. His 3% Google Translate example and electricity analogy put diffusion, systems engineering and domestic production at the center of the contest for labs, adopters and U.S. industry.
Ask SkimNews
