3 ms·
This discussion sparked another thought in me: at some point, the models might become good enough that people no longer publish things on the Web so that other
by kqr 4y ago
This discussion sparked another thought in me: at some point, the models might become good enough that people no longer publish things on the Web so that other people can search for them them and read them.
Instead, the only readers are these language models, and they whip up fresh content tailored for each new "search query", aggregating the published knowledge on the Web.
It's practically the difference between searching for "ratio sampling", getting and reading the most relevant 5 web hits, and instead searching for "ratio sampling" and having the search engine paste together the top 5000 hits into a coherent single hit that contains all the important information.
Is it better? I don't know. More convenient? Perhaps. What do I feel about it as a person publishing things on the web? Again, I don't know.
- visarga 4y agoHaving the search engine paste together the top 5000 hits into a coherent single hit opens the way for verification. You can check for contradictions.
- michaelbrave 4y agoI imagine books of research will still be written, and the model can still be trained on those. Likely, YouTube style industry news videos and comments sections will continue as well. What I really suspect will happen will be an end to end layers of devices of and systems unified in the gathering-training-giving/teaching pipeline. For example an operating system that is tightly integrated with something similar to Siri, to the point that it is the operating system, this gather's data, has conversations with users that it uses as data, asks questions to those who are experts, and then pipes that data to the larger model after some processing. This already happens somewhat with echos etc. Then the model gather's everything from books, to websites, to the aggregation of everyone's data. Maybe something like Wikipedia acts as a crowdsourcing layer of filtering for truth for it. Then it acts as an assistant in education, training, and giving answers to users much how we use search engines today but again more tightly integrated into our lives. I imagine somewhat how the computers are used in Star Trek, but more encompassing. In theory this could be as destructive towards established industries as the invention of the computer was, just needs a company who is willing to expand laterally to all aspects of data collection, processing and serving to do it, google and apple are in unique positions here, but Google would have to erode their own current offerings to do so and I have no idea how they could make money off of it.