4 ms·
Which raises the question what would be the point of LLMs when the majority of their training data is just the fluff generated by other LLMs. When the output o
by tempodox 2y ago
Which raises the question what would be the point of LLMs when the majority of their training data is just the fluff generated by other LLMs. When the output of an LLM becomes indistinguishable from its input, why go to the trouble of training it at all?
- spwa4 2y agoThe obvious answer is to do more with it. This will actually be a positive as soon as science is done by AI. As soon as the economy is done by AI. As soon as wars are fought by AI.
- nuancebydefault 2y agoI don't believe that LLM's cannot be made such that they distinguish AI from non-AI content. On the contrary. There are ways to compare similarly looking content and digging up the one that provides most insights, new ideas or actually measured data. There are ways to find out how 'old' the content is by keeping track of when it is updated and in what way (the actual data or just meta data about it, or just their interpretation or presentation). The crawlers/bots spend time and energy and hence will be optimized to not just blindly eat rehashed content. Unlike currently google's search engine. The latter feeds on the revenues of the rehashed/ad'ified content.