My take on this is that the genie is well and truly out of the bottle now, and there’s no chance of stuffing him back in there.
People have seen what generative AI can do, and - as demonstrated in this thread - are switching to, by default, asking ChatGPT, Claude, Gemini, etc… instead of doing a Google search and trawling through search results themselves. These days I use Google to search for something specific, e.g. the opening hours of a specific restaurant or the phone number of a specific company, etc… But otherwise it’s less time consuming to simply outsource tasks to Claude.
We’re all, however, still forced to use centralised, online platforms because running your own LLM, whilst technically possible, is too complicated and prohibitively expensive. But, like all things, it’ll get cheaper and easier over time. We’re still in the “early adopter” phase (even if the number of early adopters is in the hundreds of millions).
Eventually we’ll all have our own little black box, sat next to the internet router, that acts as a local LLM. We’ll no longer connect to centralised platforms and, instead, we’ll get that black box to answer our occasional questions. This may even make use of open source LLM models, so that one company can’t control what knowledge is and isn’t shared.
The centralised platforms will then become almost solely for B2B usage, performing “high performance computing” tasks that are too complex or time consuming to run locally, and where such tasks can be parallelised across tens, hundreds or thousands of agents.