
Llama 2: Open Foundation and Fine-Tuned Chat Models
Deep Papers
The Safety Benchmarks of Lama
The data set I don't know how detailed they went into the collection it was a lot of like it's open data publicly available scrubbed of sites that are known to contain PII and definitely not any of meds products. The discussion of the, the reward model that they built was fascinating because they built it over time and collected from a lot of human labelers. It speaks to the barrier of entry or the bar to entry if you wanted to get into training it's not just about having compute resources it's about having a large enough group of human beings to help align the model with the output that you want to get.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.