While I'm not an expert in AI technology, I find it hard to believe that AI development will still be going as fast as it is.
#Will AI technology development slow down in the future?
lapis plinth
junior fox
who knows 
trim prism
I would think yes, mostly hardware limits
Diffusion and transformer architectures will also meet its limit, but who knows
random wadi
We hit a triple limit atm, we have no new information for the LLMs they used everything in the Internet, Moors Law is dead around 0.7 nm is the smallest thinkable transistor we are at 3nm . The cost is also exploding OpenAi wants for O3 1000$ per request which nobody will pay.
In the future there are analoge chips both electric and photonic(1V+1V =2V) for interference which already exists in labs but making them market ready and scaling them up will take years.
On the software side something like dynamic parameterized relu function would properly also increase learning speed.(simulate hormones => ever noticed you can remeber stuff easier from shortly before going to sleep? )
In the short term only if NVIDIA or AMD would allow to link (Like NVLink) there high end cards would help normal users to make their own 100b LLMs on premise
I think NVIDIA made for interference a on premise solution . Hope this makes it clear