This is not true in my experience. Cranking out code is obviously the bottleneck, unless you have the luxury of working on a very narrow problem. The author describes a multi-modal project that does not afford this luxury.
LoRA and QLoRA are still fine tuning I thought? Just updating a subset of parameters. You are still training a base model that was pre-trained (and possibly fine tuned after).
Depends on the circumstance, and how difficult an appropriate algorithm is to write, but in my experience, if code performance is important, this tends to yield large, painful rewrites down the road.
2.5-3 micrograms per deciliter is nothing. Wouldn’t even come up as elevated by current childhood screening guidelines. I very much doubt 2-3 IQ point difference.
This is not what premature optimization is the root of all evil means. It’s a tautological indictment of doing unnecessary things. It’s not in support of making obviously naive algorithms. And if it were it wouldn’t be a statement worth focusing on.
As the point of the article is to see if Claude can write better code from further prompting so it is completely appropriate to “optimize” a single implementation.
I read it as “most people don’t connect put 2 and 2 together - the doctor could be the boys father”. It is poorly worded, but not strictly wrong I think?
It makes sense to anyone who has tried to build open source software as a sustainable business.
It doesn’t make sense to people who contribute small hobby projects or work as researchers / are government funded, which means ultimately paid for by businesses and consumers thereof.
In reality open source is very difficult to build a business around, which means that software can’t exist long term. It’s not about not wanting to be open source, it’s about realising that you and your employees livelihoods are being abused by people who see open source and take it to mean they shouldn’t pay. Especially egregious when your competitors take your work and build a closed sourced business around it.