when all you got is a tegra you just gotta do less things, you know [clicks tongue] its a shame

look at that chooch now, whooo boy

nuked it, sorry-- again.

i know, bad take, i'm sorry

I love this frame and the concept of 'optimizing for drama' so very much

i completely agree with your sentiment here; it is true in the opposite case! my point is more broadly around applications of expert trained specialists, not broad ranged "its like if we had robot slaves" my idea isn't that research is THE solution, it's the avenue i'm using to /better understand/

I'm hopeful that all of the SOTA research around small, useful, expert-level implementations for ML doesn't peter out when the LLM bubble goes boop, you know? there's so many good wins here that -- yes, unfortunately -- enabled abuses in the corporate system

It sucks ass that anything that quacks like LLMs is brigaded to shit, sometimes. I get the anger, I get the frustration. Ignorance of things isn’t a solution though. Gotta say no with math and data, too

I’m writing and training a two-headed NER on top of roberta-v3 with custom weights and annotations.

a couple of years in the '15-'18 timeline GDC 'off-moscone' was this sorta vibe. Hanging in the park, too. The whole park is just bananas. we need so much more openness in the industry

Yes!! share what you make!!

the mother of all tolet paper rolls.

computer, show me how data is time sensitive without really saying anything (the pandemic shortage -> net result of corporate planning that takes five goddamn years to shit this turd out)

roughly;
- panel ~75/100 (i don't want a crap panel)
- printing enclosure, prep, polish (+- $30)
- firmware, integration, build steps, shipping and overhead (~??~??)

If i can swing it at 170/unit, that seems feasible. probably less?

i'm going to lose whats left of my mind if this is real