This is exactly the reason why Google tried to prevent launching their LLM-based conversational search; because website owners would be all over them trying to further the narrative about how big tech is bad because their business model is at stake.
I’d be interested to know whether Bard has been tuned (or source prompted) into speaking in the first person in this way, and thus there is potentially an element of culpability here since that style of output does tend towards potential plagiarism. (“I did this / I did that / etc”)<p>Or equally this could simply be a byproduct of language completion, since the phrase “I tested the hardware” etc is likely present verbatim in the source training material.
It'll be interesting to see where this leads to. Probably a much worse web and less accurate information. As soon as online info became a source of income through advertisements counted by clicks, clicks increased and the info quality decreased. As soon as presenting search results in a certain way became a "real" business, the searches increased but the quality of the results decreased. Now that simply having access to information will create monetary value for the middlemen by extracting it from the source, more and more garbage will be fed to the LLMs but since they're not getting paid for it or getting any recognition, the <i>real people</i> behind the scenes who write the articles and produce the primary information will stop doing so. Just like the independent journalists, bloggers and hobbyists who essentially stopped having an online presence gradually over the last decade.
Sure, these AIs have decades upon decades of information available to them, but what about the new stuff? Why publish an article when nobody but Bard and ChatGPT will read them? The answer is that people won't, or if they do, they'll know it will be used to direct the AIs answers and tweak them accordingly to fool it and get the behavioural result they desire from it's users.
This also exposes an underlying problem with these tools, they only know text on the internet, and they believe everything. It's like asking a gullible child to read you the first page results and give you an answer. That's just not going to work.
I'm curious to what chatGPT would produce when presented with the same question. Obviously it would be presenting scraped results, but would it give attribution and admit it plagiarized when asked?