3 ms·
It's not hard to imagine a scenario where it remains very expensive to train and run these LLMs and the engineering breakthroughs needed to solve the known prob
by o_nate 3y ago
It's not hard to imagine a scenario where it remains very expensive to train and run these LLMs and the engineering breakthroughs needed to solve the known problems are slow in coming, and so we enter a period of stagnation. I would say this is a more likely scenario than the one in which we are standing on the verge of a new era of super-intelligent AI.
- wg0 3y agoSome conservatism is warranted here. Even if GPT-4 is downloadable, most people cannot run it unless you have pretty beefy graphics card which means we basically are talking about PCs. Laptops are out of league. We are already around 3nm landmark, GPUs are reaching up to 400 watts, 3D chiplets might add some life to scaling but I think budget now is limited. It isn't 1990. What I anticipate here is that when enough of the population has generated enough of text/images out of these, people would get an intuitive feel for AI generated content and it would not get as much traction which would push down such content producers further leaving them no choice but to either really care about creation process or get out of the arena.
- hammyhavoc 3y agoThe cost is low, and if it generates value then it's a no-brainer. I've yet to see it generate any value in my own workflows. It's just a waste of time thus far.