A few of our practical takeaways from Robotic Online Intelligence (ROI) experience in 2023, the year of LLMs and Generative AI, from a perspective of B2B data operations, research, and investment businesses:
-Data extraction, summarization, or translation applied to well-defined text (e.g. company announcements) can work well, though it still needs a human in the loop.
-Research document handling is not quite there yet, but carries a great promise for 2024, esp. with Retrieval Augmented Generation (RAG) techniques.
-The pace of change is so rapid (e.g. 4k vs 200k token limits, multimodality). Unless you are building the foundational models, the most practical mindset is to generalize the workflow and keep the flexibility to plug in and out the 'proven best latest'.
-The LLMs are still 'just' one component. They can serve as a massive booster to automation and processing of intelligence but that still needs a whole non-AI system.
And a short video with what my R2C2 version says...