Tech companies move to open AI models (blog.pragmaticengineer.com)

Recent developments from major tech companies highlight a significant shift towards the utilization of open AI models, aiming to curb soaring AI costs. Companies like Uber and AT&T reported substantial savings after transitioning to cheaper open-source models. Uber achieved a 50% reduction in AI costs by implementing strategies such as optimized model selection, ongoing benchmarking, and leveraging open weight models on inference services. This shift allowed them to exert better control over expenditures while maintaining output quality. Similarly, AT&T reported a 56% decrease in AI spending with only a 2% decline in output quality by switching from proprietary models to open alternatives. This trend signifies a broader movement within the AI/ML community towards prioritizing cost-efficiency without compromising performance, especially as companies face pressure to justify AI expenditures. Pinterest further emphasized this trend, noting that its transition to open models resulted in over 90% savings on transaction costs. The overarching message is clear: as open models become more capable and less expensive compared to their proprietary counterparts, industry players are increasingly optimizing their AI strategies to achieve better financial outcomes while continuing to harness the power of AI.