Until governments put money to openly train open source models from trusted and vetted data, this is the best we have.
Government should put money and resources like for basic research into LLMs that can be the base for model of public interest
this post was submitted on 24 Aug 2023
15 points (100.0% liked)
Quark's
1108 readers
71 users here now
Come to Quark’s, Quark’s is Fun!
General off-topic chat for the crew of startrek.website. Trek-adjacent discussions, other sci-fi television, navigating the Fediverse, server meta (within reason), selling expired cases of Yamok sauce, it’s all fair game.
founded 1 year ago
MODERATORS
This is an excellent point.
Training AIs on quality unbiased data should be looked at as being a pubic good in the classic economic sense.
Governments would find it difficult to justify paying for training an AI for a specific government need or program at that cost. To do it to establish a public resource is something different. Much more akin to why countries have statistical agencies and structured data repositories.
Source available == open source
The licenses governing use of these AIs and the fact that no one besides major corps have the hardware and manpower to train/tweak these massive AIs means they have and maintain an upper hand in making and selling these AI platforms.
Interesting article. Thanks OP for posting.
It’s notable that even most major government departments or agencies in the OECD would find it difficult to justify the resources, 10s or 100s of millions of dollars, required to train an AI, even if they have sufficient staff with the expertise. That means these not-so-open source AIs aren’t accessible.