Staff Research Scientist/Engineer @ ServiceNow | Anomaly Detection, Dynamic Translation, R&D
Hot AI take #1: The next great move in AI will not be an improvement in the power of the LLM itself. There is already evidence that “bigger is better” is meeting its limitations — smaller models are being trained which are matching the capabilities of big models. Properly created ML anomaly detection pipelines are a great example of that. The next great move is properly wrapping the LLM in ways that treat it like a child. ReAct and similar wrappers which give an LLM a limited set of options are going to provide a great tool set in the near future. If you feel the urge to think of LLMs as thinking, then view them like the 6 year old you left unattended in a kitchen and being surprised they can’t cook. #AI #MachineLearning #Innovation