Like I said above, I think there will be some very specific cases where running an LLM makes sense. Think along the lines of querying lakes of scientific data. This won’t be something of value to the average Joe or the average company. My experience with using LLMs for general search has been very hit or miss. I always end up using a general search engine, even for internal data sources.
It will if it is too expensive to generate a positive return on investment. That’s not to say that there aren’t (very) specific use cases where running an LLM makes sense, but the general use case of attempting to replace human workers with AI will likely be recognized for the pipe dream that it is. Companies are already realizing this with the AI bills coming due now, and this attitude will compound once they start having major failures and no one is still around who can fix the problems.
Those millions of people using AI for general search are currently getting it for free, which is going to end when the bubble bursts and the companies providing it go under, if not sooner when said companies try to make a profit. This is exactly what’s happening to corporate AI customers now. They are realizing there isn’t a good ROI now that OpenAI and other AI providers are being forced to charge proper costs for queries and tokens, rather than the teaser or “adoption” rates they charged before.