"... we were all better off when a reporter was someone with some hustle, deep connections in a community, and an ability to sniff out and piece together stories — especially stories that powerful people don’t want told. People who have those qualities should be able to be reporters regardless of whether they can even spell, much less reliably manifest 800 words in a classic inverse pyramid ..."
#JonStokes, 2023
https://www.jonstokes.com/p/heres-how-generative-ai-can-fix-the
(1/3)
I was sceptical about everything in this piece up to this point, but what Jon says here makes a lot of sense. Outside of my own blogs, I've only had long form pieces published on shared blog sites, where the editorial work is just as minimal as he describes. I once accidentally emailed a clearly half-written piece to the editor of one site, and only realised I'd done so when it appeared on the site.
It makes sense that newsrooms would try to cope with dropping revenues by going this way.
(2/3)
However, until such time as we can compile and run our own #MOLE on our own hardware, completely independent of the entities that developed the code and weights for them, they remain toys. Just like Free Code Linux and *BSD kernels were until they could be independently compiled on commodity hardware and OS.
Unless and until we get to that place, increased use of Trained MOLEs just hands over more of our power to whoever controls the hardware they're "trained" and run on.
(3/3)
Can't help but think this is a classic case of a solution in search of a problem;
"I’ll want a Google Docs-like real-time editor interface where I can collaborate with other writers and with the LLM."
#JonStokes, 2023
https://www.jonstokes.com/p/heres-how-generative-ai-can-fix-the
I was using an interface like that 20 years ago, without the LLM, to collaboratively edit feature articles for the Aotearoa Indymedia front page with the rest of the team. An early version of Medium had a similar collaborative drafting system.
(1/2)
Having other team members give feedback on stories as they evolved using a wiki, and co-editing the draft within the publishing interface of the website, was tremendously helpful. But I'm not convinced that adding a Trained #MOLE to the collaboration process would add anything other than overhead.
(2/2)
"I can easily imagine tools that take prompts like 'educate all the quotes', 'check all the block quotes for source links', 'make sure all the links in the piece work and are correct', add a features list widget by pulling specs from this press kit I just uploaded', 'check the photos for attribution', and turn them into staged changes that can be accepted or rejected after a brief preview or diff screen."
#JonStokes, 2023
https://www.jonstokes.com/p/heres-how-generative-ai-can-fix-the
Do we really need "#AI" to automate this stuff?
Jon has successfully found uses for his pet solution at every stage of the content-generation pipeline, and concluded that;
"... a set of tools like I’ve described here would make 'the news' more democratic by bringing this critical sense-making and storytelling work within reach of many more people ..."
#JonStokes, 2023
https://www.jonstokes.com/p/heres-how-generative-ai-can-fix-the
What he hasn't done is mount a convincing argument for how this would lead to ...
"... a higher level of quality and on a smaller scale ..."
(1/2)
I'm not even convinced he really thinks that matters. The scare quotes he puts around "the news" suggest he doesn't really think newsrooms do anything fundamentally different or more important than any other "content shop". Which is a bit like failing to see what makes peer-reviewed academic journals different from industry-funded reputation laundering blogs.
"AI" isn't nearly as much of a threat to journalism as this lack of interest in its unique value as a media practice.
(2/2)
Strypey