from Hacker News

Letter from Senators to Zuckerberg Regarding Llama Leak [pdf]

by rahmeero on 6/14/23, 3:52 AM with 2 comments

• by skeaker on 6/14/23, 4:33 AM

  I'm no fan of Meta, but a lot of these questions seem unfair. They open by saying Meta let people download the model "rather than centralizing and restricting access to the underlying data, software, and model," which is true, but then they go on to essentially blame solely this act for enabling basically every possible malicious thing that one can do with an LLM. They also throw in an unsolicited jab at Stability because of deepfakes made with SD. (In fairness they do also mention wrongs enabled by OpenAI, though those are given much less emphasis.)

  My point is, they seem to be strongly implying that they are heavily opposed to decentralization of these tools overall, which is a bummer. Decentralizing these models shouldn't be a crime. The cat was out of the bag as soon as research papers on LLMs were made public, and malicious actions were already possible with centralized OpenAI models. I don't blame Meta or Stability for the actions of their users.