Where's that figure comning from? Last time I checked (could be the 3.6 Qwen 27b) single token needed 32kb

- It's just an Independent Security Researcher.

- So that's it? You will do no action?

- Correct

I have no idea, why in that context, the number of attention heads isn't mentioned. Models have a limited set of them and obviously, a model can focus at N max things at a time, which has to put an upper bound of long context support in some way. There's just more things to lose focus to (or, mismanage the limited attention heads resources - per token)

It's a single git project at my $USER home, that is referenced in global memory. It contains as much information about work things, as possible, to be productive.

I found that, if I allowed Claude to create the notes, it actually became more and more useful, but without the guideline, I just could barely get through reading it manually.

I'd never publish anything with such origin.

I'm asking to scan projects on gitlab, go through some docs to find more grounding material, write a subarticle (in the same style), scan logs on the test env, issue some curls, etc.; until the whole article is digestible - in the "backing knowledge graph" department.

I used "CD era MSDN reference and Raymond Chen blogging style" as a starting prompt for the styleguide and my work ability to digest AI plans raised a lot.

Couldn't recommend it more. Humble, insightful and respecting the reader

One for lay-offs, because it was the best move at the time with the knowledge they had.

Second for quick correction, ability to pivot and execute quickly.

It's been always like that

I'm pretty sure it's mostly due to the training data quality. No idea, why this never gets mentioned in those discussions.

It was obvious right from the get go, that the scaling law just enabled some abilities, that were described by the underlying data and allowing the ANN to abstract it in the latent space.