From what I’ve seen chatgpt is able to generate a list of links for a topic you are discussing. They seemed very relevant to me when I tried this.

Obviously it won’t have anything post 2021 but for some subjects it’s fine.

The article states that they got a demo where they were able to ask questions about the new Bing launch, so the delay is ~hours now, not 2021 anymore (at least, for this new Microsoft thing).

There will still be links, but I'm guessing it will be ChatGPT saying: "Here's a link on the topic: ..." And if you still need more links, there will of course be a way to access the old search, but it won't be front and center.

If I type "wapo" into Google or Bing, I don't want an earfull from a LLM about what "wapo" might mean (especially if streamed slowly a la current ChatGPT), I want a link to Washington Post's current frontpage.

That seems like a trivial matter of fine-tuning the model so that it includes the link on those types of queries. For example, if you search for a website, I doubt it would be hard to have the AI infer that it's a website and include a link to it.

I mean yes, but what stopped ChatGPT2 from doing this for you?

"As a language model, I don't have the knowledge to answer your question 'who won 2024 world cup?'. These are the search result from Bing engine: ..."

Then you never need to actually access google.com or bing.com.

Yes, this is a serious concern. These GPT's have stolen the collective knowledge these places have created, particularly Wikipedia - MS will refactor it and spew it out through a Microsoft portal. In effect, ChatGPT becomes a new window into the 'hand' generated Internet - eventually this window will be a new corporate garden wall. Huge swathes of traffic will be pulled from the likes of Wikipedia, StackOverflow, blogs, etc - basically any public facing source of knowledge/information.

It's not good.

If websites don't want their content republished, they should use appropriate licenses restricting such uses.

For example, Wikipedia licenses all encyclopedic text under CC BY-SA 3.0[1] and GFDL[2], both of which permit free-as-in-libre redistribution.

[1]: https://en.wikipedia.org/wiki/Wikipedia:Text_of_the_Creative...

[2]: https://en.wikipedia.org/wiki/Wikipedia:Text_of_the_GNU_Free...

But GPT’s can restructure copyrighted text to make it different enough for it not to be direct theft. It won’t be long till you can get an entire book, feed it in and prompt it to regenerate the book in a different style. Then what happens? Or what about taking books out of copyright, then prompting regeneration in a different style with a slightly modified plot?

If you can prove there was a source and that source was used without appropriate licensing, that is a valid case of copyright infringement and possibly even IP theft.

We really need to stop muddying the water surrounding "AI" and their use of source materials. Those so-called "AIs" are just software like any other program, they are tools like any other computer application, and the copyright and licensing legal precedences that apply to software apply to them.

Their ability to transform things (art, text, whatever) at incredible speeds, means chasing copyright is going to be a game of Whac-A-Mole. Also, what percentage of transformation of copyrighted works is still infringement? A GPT can ingest a 500 word piece of copyrighted text and either summarise it down to 50 words or triple the word count in a stylistically unique manner.

Obligatory IANAL.

If you use any copyrighted source material without appropriate licensing (note: Fair Use is a form of licensing) in any way, that's copyright infringement. This is very simple and the legal world has demonstrated that fact time and time again long before computers ever came onto the scene.

Some "AI" drawing a picture derived from copyrighted materials is no different from a game using copyrighted assets from another game. If there is no appropriate licensing, it's copyright infringement.

This is akin to saying that licensing your code under the GPL will prevent GitHub's copilot from recreating it in a proprietary project. I wish it were true, but it's not.

A license won't stop others from stealing your stuff, but you will have legal standing to go after them when they do if you are so inclined.

If it works like the other GPT search engines that have popped up, it does provide links.