"AI Efficiency Boosters Face Limits: A widely used technique to make AI models more efficient, quantization, has limitations and trade-offs. Researchers found that quantized models perform worse if trained on large datasets, potentially affecting AI companies training massive models. The study suggests that lower precision may not be desirable and 'there's no free lunch' when reducing inference costs."

Source: https://techcrunch.com/2024/11/17/a-popular-technique-to-make-ai-more-efficient-has-drawbacks/