@skalskip92 could you use Astra to generate data to train an OSS model to ReID on the same 16 frame batches? since it seems to be really good you could probably create the datasets efficiently
The Nakba never ended. The genocide of Palestinians continues as the Israeli military escalates violence in Gaza, the West Bank, and Lebanon.
I'm joining @RepRashida in co-sponsoring a resolution to recognize the 78th anniversary of the Nakba and Palestinian refugees’ rights.
@sudoingX forked nanobot, switched out litellm for standard llama.cpp integration.
interested in doing something similar with aider but nanobot felt more hackable
it should also be "fucking obvious" that the rate of technical debt a team of humans can add to a codebase is much lower than that of a team of agents.
humans will eventually fix some of that debt, due to the self-inflicted pain. agents feel no such pain.
Legitimately, a math problem that might take you 10 years might take Terence Tao 10 seconds. But it's ok, there are billions of math problems, more than Tao can solve in a lifetime, so if you like math your job is just to find one Tao happens not to be interested in
I really do believe this btw. Every time I felt really frustrated with a concept, the problem was always, every single time, resolved via familiarity. It's so true and so well known that competition math and undergrad logic books will often explicitly say it in their introduction
the filesystem is just a graph database.
All of you saying you don't need a graph for agents while using the filesystem are just in denial about using a graph.
I don’t know why so many people are obsessed with the idea that it’s desireable to maintain software while having zero technical understanding of any part of how it’s built.
It’s just not efficient to prompt ai with “fix x”, if you don’t even understand how x works.