This is absolutely huge. LLaMA weights leaking was a big deal, but you couldn’t actually use them without attracting Meta’s ire. Would love to see some benchmarking vs. LLaMA and GPT.
It's not clear how this process applies to model weights. Once you run another training epoch on them, the data has changed. What is the essential copyrightable, trademarkable or patentable thing that remains? A legally untested question for sure.
"Ire" is a synonym for "anger" or "wrath"
It’s not an acronym.