Things are starting to look very real!
- See if I can get the OAuth version of the CSE working – turns out there isn’t one – it’s all V1. However, Google now charges for every custom search over $100/day. Can get expensive very quickly
- 4:00 Meeting
- 10:00 Meeting with Aaron and Rukan
- Cancelled Quantico meeting
- Had a long discussion with Rukan about how to work with FOM and trajectory information
- Work on Resilience/Robustness section referencing this: the-many-faces-of-resilience. I want to better describe resilience and robustness using network terms and show how war elephants, when used properly, were part of a resilient network including nodes for the elephant, the mahout, and the soldiers.
- Compute-Optimal LLMs: This tool helps you find the optimal LLM given your compute budget. Alternatively, you can also use it to find the optimal compute budget for a planned model. The tool is highly based on the work of Kaplan et. al (2020), Hoffman et. al (2022), and most recently with new results on varying modalities, Aghajanyan et. al (2023). I do not make any claims about the accuracy of these predictions, but merely provide a tool to try to easily calculate what these predictions actually mean for your model.