DougBTX

For numerical work, there’s no reason to assume the computation is happening in the LLM. All frontier models could handily write a Python script to do linear regression then report the result, so the quality of the responses here depends entirely on the prompt + harness.

> we set up limits for a maximum of 256 process per UID, and regularly get folks asking "what does this 'cannot fork' message mean?"

The max limit on 64 bit systems is what, 4,194,303? So if you have over 16,000 users per VM this limit makes sense, otherwise it just seems user-hostile.

April last year

Yeah, we’re in a funny position. By all accounts it is fair use (at least in the US) to train models (and build search indexes, e.g. Google Books), but sharing the books dataset itself is absolutely forbidden (clear non-transformative copying).

Anyone that wants to train a model needs to procure and destroy their own physical copy of each book!

Oh, thanks, that’s interesting, I thought it covered that too!

The weights are code, the prompt is code, the output is code.

Is the meat code?

The train/test split is one of the fundamental building blocks of current generation models, so they’re assuming familiarity with that.

At a high level, training takes in training data and produces model weights, and “test time” takes model weights and a prompt to produce output. Every end user has the same model weights, but different prompts. They’re saying that the constitution goes into the training data, while CLAUDE.md goes into the prompt.

I can’t tell if this is sarcasm or not! Radio buttons support keyboard navigation without JS.

Spicy:

> Use caution when using the manual door release; the window will not automatically lower when the door is opened and damage to the window or vehicle trim may occur.

Manually opening the rear doors is a destructive operation!

Yes, from the article:

> To support the cluster’s massive scale, we relied on a proprietary key-value store based on Google’s Spanner distributed database... We didn’t witness any bottlenecks with respect to the new storage system and it showed no signs of it not being able to support higher scales.