Post ATmDAfZvq2ZLiDzNIW by ayourtch@mstdn.io
 (DIR) More posts by ayourtch@mstdn.io
 (DIR) Post #ATlsaNgIu80MhGO8ES by ricmac@mastodon.social
       2023-03-19T09:42:14Z
       
       0 likes, 0 repeats
       
       Couple of ChatGPT things stood out for me this morning (unfortunately both were found on twitter, but that’s just the way it is right now). First, this observation from Rafat Ali. The replies are fascinating: TLDR, he thinks the AI is hallucinating summaries of paywalled news stories. Wow, that is not great for media *consumers* (ie all of us). How journalism responds will be intriguing. You have to think some kind of revenue share between the source (media) & the summarizers (AI cos) is needed
       
 (DIR) Post #ATlsaOAn4mOKDoeTke by danyork@mastodon.social
       2023-03-19T09:50:13Z
       
       0 likes, 0 repeats
       
       @ricmac oh, interesting. I will have to try that next time I hit a paywall.
       
 (DIR) Post #ATlsaOh38qCBprkF28 by bortzmeyer@mastodon.gougere.fr
       2023-03-19T10:56:39Z
       
       0 likes, 0 repeats
       
       @danyork @ricmac If a summary created from nothing (or from just headline and URL) is convincing, it may mean that the article itself was quite poor in content...(ChatGPT is really strong against producers of "fast content" - the equivalent of "fast food".)
       
 (DIR) Post #ATmC5x5RDSdcbvucXw by simon@fedi.simonwillison.net
       2023-03-19T14:33:09Z
       
       0 likes, 0 repeats
       
       @ricmac I wrote about that here - it's an incredibly common misconception that ChatGPT can read content from a URL that you pass it... because ChatGPT convincingly pretends that it can! https://simonwillison.net/2023/Mar/10/chatgpt-internet-access/
       
 (DIR) Post #ATmCMuvYUvoSpvNaCG by ayourtch@mstdn.io
       2023-03-19T10:14:39Z
       
       0 likes, 0 repeats
       
       @danyork @ricmac you don’t even need a paywall: I just tried http://stdio.be/blog/2015-02-20-An-Experiment-In-Overcomplication/ with ChatGPT and you can check the results… not so stellar… it looks like the URL leaks enough information that the ChatGPT is able to construct a plausible-looking story based on its generic knowledge on the subject ? (And if so - why would one demand a revenue share ? Technically, GPT just took the openly available information and combined it. Much like you and me could do.)
       
 (DIR) Post #ATmCMvYu8dHwnxn0Wu by ayourtch@mstdn.io
       2023-03-19T10:32:33Z
       
       0 likes, 0 repeats
       
       @danyork @ricmac and just as a second proof point, a couple of more “news” examples,  including the one originally posted. As you see it made a “summary” for “my article as well”. Does it highlight how absurd it is to talk about revenue sharing here ? 😂
       
 (DIR) Post #ATmCMw0uSVgqCotNBI by simon@fedi.simonwillison.net
       2023-03-19T14:34:50Z
       
       0 likes, 0 repeats
       
       @ayourtch @danyork @ricmac really important to get the message out that it CANNOT read URLs that you give it! It's a terrible usability flaw that it pretends that it can https://simonwillison.net/2023/Mar/10/chatgpt-internet-access/
       
 (DIR) Post #ATmDAfZvq2ZLiDzNIW by ayourtch@mstdn.io
       2023-03-19T14:45:11Z
       
       0 likes, 0 repeats
       
       @simon @danyork @ricmac the fact that journalists fall for it highlights one of the issues with today’s journalism: trend to sensationalism and lack of critical fact checking. Anyway I just had an epiphany (see my other post I just made): LLMs are a perfect infinite source of emotional crack cocaine. The People don’t care about the facts; they care for the adrenaline kick the text gives to them. LLM will be able to make that with surgical accuracy.
       
 (DIR) Post #ATnA6qVX9nRbkHNQDQ by danyork@mastodon.social
       2023-03-20T01:45:52Z
       
       0 likes, 0 repeats
       
       @simon @ayourtch @ricmac Fascinating! THANK YOU for writing that post and sharing the info!