Running Llama locally is the only interaction with AI I am comfortable with. And it's a great experience. We only need simpler ways to train the models further on local data, a method that can be deployed by the normal user without much tech involved. I think a better/easier training framework is more important than a new model.

👍

Tis

I think it will be better to live in a world where AI is widely deployed so that smaller actors can check the power of larger bad actors.

Wish list just dropped. localisation is great. A lot of corporations are rushing towards llm adoption through microsoft (their localisation is a few tens of a thousand annually if i am not mistaken). Reminded me of your italian talk on why the giants keep getting bigger and the danger to it. I guess one step further might be in isolating ownership, esp for customers who leverage the tech, maybe end to end encryption or de-identifying data idk. It might prevent customer’s data leakage or for the business owner to be put in a position to hand it to the govt when they come knocking.

Did Zuck drop acid?

he is commoditizing the layer to hurt others. 3 other big player probably spent hundreds of millions. by spending a smaller amount he is hurting 3 big ones. ultimately small players gain. i may switch to 3.1 from 3. right now Ostrich 70B is based on 3. but i have to first check how much disinformation is in 3.1. for that i have to check truthfulQA and also run some tests. need i say truthfulQA is not that truthful 🤦‍♂️

I'm going to take your word for it

FOSS now and then. View quoted note →