scaling
The thesis in AI that increasing model size, data, and compute predictably improves performance.
As defined in Michael Nielsen
Other definitions
The technical challenge of increasing blockchain transaction throughput and speed while maintaining security and decentralization; addressed through L2 solutions, sidechains, or increasing block size
90 - 5 Mental Models for Web3 | Chris Dixon
Where it comes up · 11
- Artificial General Intelligence's Five Hard National Security Problems4 claims
- The Disruptive Potential of GenAI for Emerging Markets | Rem Koning | Leading with AI4 claims
- 362 | Luis Bettencourt on the Universal Properties of Self-Organizing Cities4 claims
- The future of intelligence | Demis Hassabis (Co-founder and CEO of DeepMind)3 claims
- Audrey Tang2 claims
- Esther Dyson on the Attention Economy and the Quantification of Everything2 claims
- Michael Nielsen1 claim
- Dwarkesh Patel Podcast - The next big breakthrough will be AIs learning on the job1 claim
and 3 more
Most important claims · 10 of 24Search all claims for “scaling” →
- Changing behavior permanently is very different from changing it during an active intervention, and almost nothing scales—both the durability of behavior change after a program ends and the failure of interventions to scale are enormous, recurring problems in the field.Susan Mayer on What Money Can't Buy
- On the technical side it would be weird if AI scaling just stopped working, so scaling will likely continue; the more underrated barrier is on the knowledge side, where the universe may be fundamentally illegible, meaning there may be no fruitful way to apply more intelligence to certain frontier problems like unifying general relativity and quantum mechanics.Tyler Cowen — Hayek, Keynes, & Smith on AI, animal spirits, anarchy, & growth
- People walk faster in larger cities because space can be compressed by density but time cannot, so to fit more interactions into a fixed day they must move faster between them, which is the reason pedestrian speed rises with city size.362 | Luis Bettencourt on the Universal Properties of Self-Organizing Cities
- Skin in the game engages with two aspects of social thought usually neglected: dynamics (how systems change over time) and scaling (how rules and behaviors must change qualitatively as group size increases); these are mathematical transformations, not linear adjustments.Skin in the Game | Nassim Nicholas Taleb | Talks at Google
- Starting a company successfully is easy; the hard part that creates real value is growing it, scaling it, and making it sustainable when the founder leaves—so the current veneration of the hero-entrepreneur is excessive and undervalues the middle management and expanding teams that actually make people productive and let one team's success be emulated by others.Esther Dyson on the Attention Economy and the Quantification of Everything
- When a city's population doubles, average per-capita quantities such as wages, GDP per capita, and expenditures rise by roughly 16 percent, so larger cities produce and spend more per person in a continuous superlinear pattern.362 | Luis Bettencourt on the Universal Properties of Self-Organizing Cities
- The widely believed theory that scaling is very important is a capital-friendly story that shifts negotiating power away from individual researchers toward centers of capital, while it remains 'just a story'; combined with big companies publishing fewer papers, this damages individual researchers' ability to build public brands and capture value.Michael Nielsen
- The speaker created a minimal viable version ('minimum viable AI') as a deliberate design choice, deliberately testing effectiveness before scaling, as a responsible approach to deployment that avoids accidentally harming businesses through poorly-tested tools.The Disruptive Potential of GenAI for Emerging Markets | Rem Koning | Leading with AI
- What is currently being observed is not a monopoly situation in frontier model development; instead, there are multiple actors in the frontier model development layer, though other layers (hyperscalers, chips, machines used to make chips) have different market dynamics, with future monopolies in frontier model development dependent on whether scaling requirements continue and whether frontier model pre-training remains easily transferable across tech stacks.Artificial General Intelligence's Five Hard National Security Problems
- If AGI emerges soon under a scaling paradigm requiring large compute investment, proliferation control may be possible by limiting access to supercomputers used for training and inference, but this assumes a particular technical paradigm and timeline; AGI might instead emerge through algorithmic innovation, or as aggregation of specialized models, or distributed across many models, all of which would proliferate differently.Artificial General Intelligence's Five Hard National Security Problems