How do you work with generative AI models? What positive outcomes have you seen from generative AI adoption? Model preferences continue to usage of closed-source models Similar to last year, 61% of orga- Processes like RAG and fine-tuning Proprietary data is a key ingredient evolve and remain a key decision easier. Many closed-source models nizations stated improved oper- introduce the complexity of integrat- to power performance enhance- for an organization’s AI strategy. also outperform open source ational efficiency as the leading ing external data sources in real- ments for generative AI models. The largest increase in usage came models out-of-the-box. Despite driver behind adopting generative time, ensuring the relevance and While Scale’s machine learning from closed-source models with that, open-source model usage still AI. Improved customer experience accuracy of retrieved information, team proved how fine-tuning can 86% of organizations using these increased from 41% to 66%. This is came in second at 55%. managing additional computation- enhance model capabilities, 41% of models compared to 37% the year likely due to the flexibility open- al costs, and addressing potential organizations lack the ML expertise prior. This is likely due to a combi- source models provide for fine-tun- Despite growing adoption, there are biases or errors. Fine-tuning requires to execute the data transformations nation of factors. Many organiza- ing and hosting. The smallest still a number of challenges that stall careful selection of data to avoid and measure and evaluate results to tions have existing contracts with change in model preferences were widespread use of generative AI. 61% overfitting and ensuring models justify the initial investment. cloud service providers who in turn organizations that trained their of respondents cited infrastructure, remain generalizable to new, unseen have partnerships with closed- own models at 24% in 2024. tooling, or out-of-the-box solutions information. source model developers, making not meeting their specific needs. 08 09

AI Readiness Report 2024 - Page 10 AI Readiness Report 2024 Page 9 Page 11