Generative AI's biggest challenge? Showing ROI - here's why

Summary

Steve Jones, vice president at Capgemini, explained at a Databricks conference in San Francisco that the biggest challenge for companies deploying generative AI is proving return on investment. Proof-of-concept demonstrations are impressive but hard to monetise. He recommends rigorously testing large language models, including deliberately feeding them incorrect information, and warns against using LLMs for simple tasks that do not warrant them, which he describes as "burning money". He advises comparing the performance of several models (GPT-4 Turbo, Llama) rather than relying on a single one, and expects it to take two to five years before generative AI is widely adopted.

Summary based on the full text of the article.