"Oh we just need more compute and all the problems will go away." There are almost certainly diminishing returns on additional compute for these LLMs and stable diffusion engines, but they won't hear it.
Instead it's just "We just need more compute, just give us more compute, fire up every decommissioned powerplant in the country and steal everyone's drinking water, just a little more compute and AI will finally work!"
It's definitely starting to look like we're well into diminishing returns territory with the current LLM approach. I think GPT algorithm is a useful piece of the puzzle, but clearly some fresh ideas are needed to move forward in a meaningful way. I'm really hoping the hype will start to die down so that more effort goes into exploring other techniques instead of all the air in the room being sucked up by LLMs.
They are so desperate for AI to work out.
"Oh we just need more compute and all the problems will go away." There are almost certainly diminishing returns on additional compute for these LLMs and stable diffusion engines, but they won't hear it.
Instead it's just "We just need more compute, just give us more compute, fire up every decommissioned powerplant in the country and steal everyone's drinking water, just a little more compute and AI will finally work!"
It's definitely starting to look like we're well into diminishing returns territory with the current LLM approach. I think GPT algorithm is a useful piece of the puzzle, but clearly some fresh ideas are needed to move forward in a meaningful way. I'm really hoping the hype will start to die down so that more effort goes into exploring other techniques instead of all the air in the room being sucked up by LLMs.