Gemini keeps being rather impressive though, even their iterative updates have improvements, though I'm seeing a significant slowdown in the improvements (both quantity and in how much they improve) suggestion a wall may be approaching.

That said, technologies like this can also go through a rollercoaster pattern itself. Lots of innovation and improvement, followed by very little improvement but lots of research, which then explodes more improvements.

I think LLMs have a better chance at following that pattern than computer vision did when that hype cycle was all the rage