Interesting Links for 16-01-2023

[andrewducker]

1. Everything Everywhere All at Once — How to Shoot a Sci-Fi Kung Fu Epic (on a Budget)

(tags:movies scifi MartialArts everything )

2. Text and data mining copyright exception proposal to encourage AI focus and innovation in the UK

(tags:UK ai data copyright )

3. Google's very expensive coders produce a security vulnerability in c/c++ every thousand lines of code. If you care about security then you shouldn't use them.

(tags:security programming memory )

Comments

[helen_keeble]

If I’m reading that copyright proposal correctly, it means that it would be considered legitimate use for someone to scan/ocr/download the contents of copyrighted books into a database, train an AI on it, and produce a commercial tool for generating similar text, right? That would seem a considerable land grab.

In fact, IIRC such a database already exists - didn’t Google get into conflicts with publishers over digitising contents from copyrighted books? So it would seem legal for Google to take that massive database and train an AI to produce derivative content.

I cannot see trad publishers being happy with that AT ALL.

(There would be even greater fury from authors, but government doesn’t care about the views of people who have the temerity to attempt to make a living in the arts)

I mean, it’s exactly what’s happened to visual artists having their work taken without permission and used to train AIs, but at least there (ethical considerations aside) it was content scraped from the open internet. Making it legal to just dive right in and train AIs on commercial text (…or scanned images, ripped dvds, recordings of performances, contemporary music archives, etc) without rights holder permission would seem to be opening a whole can of worms.

[andrewducker]

Produce derivative content, yes - although it would have the same legal limitations as person-produced derivative content. It wouldn't be allowed to write a sequel to the Lord Of The Rings, for instance, even if it was competent at doing so.

[helen_keeble]

Indeed, but it would be fairly easy to produce synthesised textbooks/self help/diet books etc. Or some types of formulaic fiction (I am not dissing formulaic fiction here - I make a living writing it, which is why this whole area is of very personal interest 🙂)

Actually, I could see certain publishers like Harlequin being delighted by a commercial AI that can churn out first drafts that only require some light reworking by a ghostwriter/editor to be publisher. Let’s cut out those pesky authors entirely!

Though it won’t be a big trad publisher that does this first - it’ll be the erotica content mills switching from using poorly-paid ghosts to straight-up AI (with some human editing). I am fairly certain some of them must have done so already. But when an AI content mill starts challenging trad publishing houses in spheres like cozy mysteries or thrillers, things will get…interesting. Especially if it’s perfectly legit for the providers of the AI model to proudly proclaim that it’s been trained on, say, Dan Brown or Nora Robert’s complete works and can write in the same style.

I know of several authors who are using GPT-3 - via the Sudowrite commercial interface - to generate outlines and substantial chunks of first drafts already. Some of them are even training the AIs on their own content first so it can emulate their style. I’ve played around with using ChatGPT as a brainstorming partner myself, in fact.

(I am actually fairly sanguine about the ability of authors to continue to make a living in a world with increasingly good text AIs, but it is going to require some _considerable_ agility and willingness to adapt)

[andrewducker]

Part of me thinks that if AI can do it well enough that the readers don't care, then that's fine. Let publishers produce books-to-order, customised by whatever individual readers fancy having in their novels. It's not like the vast majority of writers ever made any money at it anyway.

But I hope it doesn't put people off of writing for fun. And I will feel bad for the few people out there who *do* make a living from it.

(And I suspect that it will be a fair while until those people are outwritten by an AI, and that once we reach that stage we will have other issues to contend with...)

[helen_keeble]

Well, I make more as a romance author than I ever did as a senior software consultant for a top consultancy firm. 😉 And I could easily name ten authors just in my narrow sub subgenre who are making a lot more than me. And that’s not even getting into the _really_ lucrative genres like dark contemporary romance or thrillers.

There’s a surprisingly large number of people who’ve been quietly making bank as indie authors since 2010ish. Text AI is potentially a serious disruption to the current industry… but thankfully, a lot of us are pretty good at pivoting as necessary.

It’s an interesting topic from an indie author perspective as we stand to be disrupted by text AI, while simultaneously being able to get big benefits from both text AIs and art AIs. Cover art/illustration is a big expense for me, and I’m severely limited by available stock photos. Which puts me in the uncomfortable position of being able to benefit greatly from AI that’s going to potentially put some of my photographer/artist friends out of business. Ouch.

[andrewducker]

I would be very interested in understanding more - have you (or anyone you know) ever written anything public about the economics of it all?

[helen_keeble]

Look up Mark Dawson, Lucy Score, Michael Anderle, Kevin J Anderson or Joanna Penn. They’re all very prominent indies who give frequent talks/interviews on the nuts and bolts of the industry. I’d put some links here but everything I’m easily pulling up is behind a newspaper paywall, alas (I know you have ways of getting around those!)

[andrewducker]

Cheers!

I'm mostly going on things that I've read about the income of authors like this which says that:

The median earnings for primary occupation authors (writers who spend more than half their working time writing) are £10,497 a year...
There is considerable inequality of earning power amongst authors, with the highest-earning 10% of writers taking home about 70% of total earnings in the profession. Meanwhile, most writers need a second job to survive, with just 28% of respondents making a living from writing alone without a second job.

Obviously there are some authors who do vastly better than that. I suspect there's a power law curve to earnings, as there usually is to such things.

[helen_keeble]

Yeah, Mark Dawson did a somewhat bewildered reply to that, but it was in a private author group so I can’t link it to you directly. As far as anyone can determine, that study was either only looking at trad pub authors, or gathered data from a very skewed sampling of indies (since they definitely didn’t approach the places where the pros hang out!).

There was another researcher who was specifically surveying indies (I filled in the survey myself) but I’m not sure whether the results have been published yet. I’ll check…

Amazon’s own official figure is that there are over 1,000 authors making more than $100,000 per annum just through Amazon self publishing. We actually think that’s _considerably_ under reporting.

Of course, the overall figures are skewed because there are so very many people who _want_ to be an author, and even a lot who write full time while being supported by some other income source.

[andrewducker]

Cheers, that makes sense. I'm glad that more authors are making a living than I thought!

[helen_keeble]

Actually, the real mess is going to come when (if?) different countries start to diverge in the legality of what AIs can and cannot be trained on. If UK law says you can use copyrighted text and US law says you can’t, and I use a UK-based AI to write a book… can I still publish it in the US? If not, how much would I have to change to turn it into legit human-authored work?

The legal cases are going to be almost as fascinating as the technology itself.

[simont]

The legal cases are going to be almost as fascinating as the technology itself

... briefly, until judges get the idea of having the same AI generate their judgment writeups :-)

[andrewducker]

Absolutely. I shall watch from a safe distance.

Oh, and legality can already vary from country to country - Winnie The Pooh is in copyright in the UK for another 4 years, for instance...

[calimac]

It'd be interesting to ask the AI to write a sequel to Lord of the Rings, just to see what would happen.

I'm hoping it would tell me, "I'm sorry, Dave, I can't do that."

[mellowtigger]

#1 Very good movie. I keep intending to write about it.
#2 I have too many thoughts on AI, capitalism, and the nearing Star Trek/Orville economy. I need to organize them better.
#3 It reminds me of that old alleged quote, "Democracy is the worst form of government, except for all the others." I trust Google code (even c/c++ code) as much as any other program, just not their corporation.

[andrewducker]

I should have been clearer that the "them" at the end of #3 is "C/C++" and not "Google".

[channelpenguin]

LOL. I assumed either "C/C++" or "C/C++ programmers"

[jducoeur]

3 - I mean, yes: memory management in C/C++ is a horror. I switched away from them in, what, I think 2002? It kind of astonishes me that so many applications that don't need to be down-to-the-metal have continued using them for so long. And with the rise of far better native-compiled languages like Rust and Scala Native, I can't see ever wasting my time with them again.

(NB: I spent 15 years working professionally mainly in C and then C++. My distaste for them is very well-earned.)

[andrewducker]

Yes, I'm quite glad to have not worked with them. I'm sure I would have leaked a lot of memory in my time!