Although a lot of the hype around LLMs right now will prove to be premature, there is one uncontroversial "killer app" for GPT-alikes: semantic compression. Inp

Although a lot of the hype around LLMs right now will prove to be premature, there is one uncontroversial "killer app" for GPT-alikes: semantic compression. Input a bunch of text, output summaries and answer queries about it. One example I've heard where this will kill a lot of jobs is banking: banks currently need to spend a lot of money having humans parse and develop policy changes so that the bank can remain compliant with changing legislation. I haven't had direct experience, but I've been led to believe that LLMs are already starting to automate this kind of work, and early experiments have been successful. Aside from banks, I suppose the general principle will apply to any kind of work where we currently need humans to comb through and curate a bunch of data, before regurgitating summaries of it. Historians, statisticians, librarians, accountants, lawyers, doctors and journalists come to mind. I don't claim all these fields will become obsolete per se, but I think we'll, at minimum, see them evolve into new forms. For example, I'm having trouble imagining how a human applied statistician will be able to compete with a well-prompted LLM which is able to hook into external statistical compute engines when it needs to do some actual number crunching. It doesn't mean human statisticians will no longer exist, but we may see fewer applied statisticians in industry, and the remaining statisticians will earn their crust through either software development or writing about more theoretical approaches in academia. (Despite optimistic appraisals of LLMs, I don't think either software development or academic researchers will be made obsolete within 10 years.) Aside from LLMs, to think about what technology will be obsolete in 10 years from now, it pays to look at what became obsolete in the last ten years, and why. What we overwhelmingly see is a trend where applications formerly handled by dedicated and/or analogue devices are being absorbed into web applications and smartphones, or equivalent smartphone-like technology in cars and home appliances. It stands to reason that, where it is possible to virtualize and digitize a task, it will happen soon if it hasn't happened yet. This leads me to another prediction, that TVs will no longer ship with remote controls in 10 years, because TVs will expect you to control them with smartphones.