The author suggests that the solution lies in connecting general, industry, and organizational data, despite the challenges of differing semantics, permissions, and data quality. The author argues that until data on the internet, intranets, and even offline can be interconnected by use case, AI models will not be able to provide deep, meaningful insights. The author concludes that the focus should be on sorting, organizing, interconnecting, updating, and rendering safe the data that AI models process.
Key takeaways:
- Large language models (LLMs) like ChatGPT, despite their initial hype, have not replaced human jobs as expected due to their limitations in providing meaningful and cost-effective results.
- The main issue with these models is not the volume of data they are trained on, but the type of data. They lack access to specialized data, making them less useful to organizations than anticipated.
- Attempts to solve the problem by linking LLMs to proprietary databases have not been successful due to issues such as different semantics, outdated information, and permissions.
- The solution to making foundational models more useful lies in connecting general, industry, and organizational data, ensuring they are updated, permissioned, and working together. Until then, AI models will remain limited in their utility.