LLM2LLM: Synthetic Data for Fine-Tuning (UC Berkeley)
Vložit
- čas přidán 12. 09. 2024
- LLM2LLM: Can LLM teach other LLMs new knowledge? How is it done? What is the performance of those AI systems? Can LLMs generate high quality datasets for the fine-tuning of other (smaller) LLMs? For edge devices?
All questions answered in the latest video on synthetic data generation and synthetic data augmentation.
#ai #airesearch #newtechnology
Very excited that I see a video from you on this topic 🎉
Very interesting!
I wonder: if I would like a LLM that is specialized in for example Physics knowledge, how could I use this method to generate such a specialized LLM?
One glaring omission form this list is a check of data's veracity. Or did I miss that?
An easy way to do it would be to feed the answers from the originating LLM back to itself and have them evaluated for accuracy, truthfulness.
Thanks for the video! What software do you use to create this beautiful presentations?
Is anybody using LLMs to process original source data, preparing/optimizing it for input?
little grasshopper* LLM