flâneur

Open AGI - by Paul Veradittakit - VeradiVerdict

veradiverdict.com · 1,115 words · saved by 1 readers

Incentive alignment: AI is made for people, but its development and rewards are decided in boardrooms Firstly, AI models do more and more computations (FLOPS) and have higher and higher training costs. OpenAI is on track to lose $5 billion this year due to high costs. AI companies have so much additional baggage: sales teams, legal departments, HR, distribution, purchasing, and more. How about designing infrastructure that trustlessly distributes models in a monetizable and ownable way, so that researchers can focus on building models, not on extraneous tasks? Compute Trends Across Three Eras of Machine Learning Amortized hardware and energy cost to train frontier AI models over time Secondly, decision-making is decided from the top down. The decision on the metrics to follow, market to target, data to gather, and modality to include are decided by internal hierarchy. Centralized decisions are made to benefit shareholders, not end users. Instead of predicting the use cases, why not let

Over the past few years, two new holistic problems in AI development have emerged: Resource management: AI development is not cost-scalable Incentive alignment: AI is made for people, but its development and rewards are decided in boardrooms Firstly, AI models do more and more computations (FLOPS) and have higher and higher training costs. OpenAI is on track to lose $5 billion this year due to high costs. AI companies have so much additional baggage: sales teams, legal departments, HR, distribution, purchasing, and more. How about designing infrastructure that trustlessly distributes…

related reading