AI in the "PaLM" of Your Hand
This talk explains the fundamentals of Large Language Models as they apply to Google’s PaLM 2.
First, it touches on the high-level structure of LLMs and some of the techniques Google used to train PaLM 2 so that you gain an appreciation for all the time, thought, and environment that makes them possible. After the technicalities of LLMs are explained, the talk covers the various productivity tools Google offers, particularly in Workspace and GCP, to help users be more productive with project management or coding tasks. These tools are great for those who don’t need custom models. For those envisioning custom applications, MakerSuite comes into play. Here, users will see how prompt engineering works to get the best results out of PaLM 2 out-of-the-box, then invoke the PaLM 2 API with the desired prompts. Users can even fine-tune private models based on PaLM 2 and invoke them using PaLM 2 API plus OAuth. MakerSuite allows for several tunable parameters to be customized, and these are explained thoroughly. You should come away from this talk with an:
Appreciation for the amount of brain power and compute power behind LLMs, Knowledge of Google’s AI-driven productivity tools, Understanding of how to use prompt engineering and fine-tunable parameters to make the best use of the PaLM 2 API, and how to experiment and arrive at the optimal results using MakerSuite.
