The models themselves represent the biggest deflation cases I've ever seem.
The charged cost of a frontier model is ~200x lower than 2 years ago, and the ones we are using now are much better - although measuring that and how much is challenging. Building a "better than GPT-4" model is also vastly cheaper than building GPT-4 was... perhaps 1/100th?
- voting population? What are you on about? It's everyone older than 18.
- you are not allowed to say "let's go and kill xxxyyy" or "burn hotel xxxyyy" but more or less you can say anything else. You might get sued if you say "Kier Starmer is an XXXYYY" but possibly not.
- this is using a system such as the one that operates in many countries - like France. But note: Germany ditched jury trials in 1924...
I've seen a couple of stories - like this one - https://www.dailymail.co.uk/news/article-15302729/Assault-vi... - that make me think there's a very liberal interpenetration of "hate crime", as in there doesn't need to be an underlying crime but speech alone if offensive enough is criminal.
16 year olds have just been given the vote by the Labour government in an attempt to shore up their support because they’re going to lose the next election. Lowering the power of my vote as an adult taxpayer by enfranchising teenagers who get their political nous from TikTok is pretty disgusting.
I think that Google didn't see the business case in that generation of models, and also saw significant safety concerns. If AI had been delayed by... 5 years... would the world really be a worse place?
reply