I mostly use tools like this for data exploration in Jupyter notebooks and PyVista was fine when I tried it out last year. However, I found I could get results much faster with Vedo[0].
No hate though, I'm so glad there are options in this space!
Your point aside, counting letters is not a good task for LLMs since they aren't trained on letters. They are trained on tokens which represent words or parts of words.
> Preferred by anyone who's actually using and modifying the work.
> ...fine tuning is preferred by everyone
How do you know this? Did you take a survey? When? What if preferences change or there is no consensus?
> The only people I've seen who've asserted otherwise are random commenters on the internet who don't really understand the tech.
There are lots of things that can be done with the training set that don't involve retraining the entire model from scratch. As a random example, I could perform a statistical analysis over a portion of the training set and find a series of vectors in token-space that could be used to steer the model. Something like this can be done without access to the training data, but does it work better? We don't know because it hasn't been tried yet.
But none of that really matters, because what we're discussing is the philosophy of open source. I think it's a really bad take to say that something is open source because it's in a "preferred" format.
I have a very different memory of my time on Facebook 10 or so years ago... It felt like every two weeks some update would change my settings to "public" in some way.
Perhaps my question was worded poorly. I'm saying is: the premise of the article is wrong. It attributes the 29% figure to AI but the source attributes the 29% to new datacenters. It seems unlikely to me that 100% of new datacenter usage would go to AI.
I tried to find the source of the 29% figure, but it's defined as data center usage. I understand machine learning is resource intensive but how much is it really? Are the numbers for this available to the public?