Signal Training compute for AI models doubling every three to four months

Summary

The report states that the computational resources used to train AI models are currently doubling every three to four months, citing an independent review on the future of compute from the Department for Science, Innovation and Technology. As compute has grown, AI systems have become capable of complex tasks such as generating content, passing exams and producing images from text prompts. From the 2010s, AI models turned to deep learning, finding patterns in large datasets without explicit programming. This shift let a few large technology firms with vast data and computing resources dominate the market, concentrating market power. The report also warns that wider AI use could push up energy consumption and worsen global inequality because of unequal access to and ownership of technology.