Upgrade your account to a Community Supporter account and remove most of the site ads.
Enchanted Trinkets Complete--a hardcover book containing over 500 magic items for your D&D games!

Reply to thread

Message: <blockquote data-quote="Jfdlsjfd" data-source="post: 9825850" data-attributes="member: 42856">I think maybe his methodology has a flaw as he only tried short excerpts. It's easier to get a text right when it's a paragraph long (hence the success of things like Copilot to execute tasks like "take my email draft and turn it into a fully written one"). What LLMs currently don't do well is writing long texts with consistency and getting interesting ideas, so they are better as helpers. But it's not the main point the article addresses.Thing is, even with an ability to generate only low-quaity text, the threshhold for marketability is low. I have been reading extremely nice pieces of fanfiction, some of which I judged better than the original... yet those gems stand in an ocean of very shoddy fanfiction. And this low-level fanfictions gets upvotes. So for a category of readers, "excruciatingly bad" is worth reading. If you strongly want to read about a topic, I guess you lower your expectations? One can deny that people will love all sorts of low quality content, but I think it's not representative of the market. In the RPG scene, we had lots of 3.5 products on DTRPG that were extremely bad, yet sold. And with sufficently advanced technology, the cost of making a book might make it so that it's worth doing it. If one only needs to prompt "write me a monster book statting creatures for D&D 5.5 and generate a fully designed pdf for publication" and spend a few cents in electricity, then even a single dollar in a PWYW offering will make that economically viable. And the threshhold of having one single person on Earth interested enough to throw a dollar is probably low, quality-wise. Of course, I can only see that being transitory: at some point, when anyone realize that they can prompt an LLM themselves to get the same result they will. And then, only products with more elaborated production process (like, some actually creative humans injecting their creativity to provide the added value, while letting most of the work being done by the LLM) will emerge, because there will still be a market for new ideas. It will lower the entry barreer to content publication even more than fanfiction websites did compared to trying to get your book published, and we'll probably need someone to read through all that to sort the chaff from the grain. We're not there yet, but it might have the beneficial effect of allowing authors to produce more efficiently. Like the Akutagawa prize laureate Rie Kudan, they might focus on what they are best at and let the AI do the rest (in that particular case, it was "imitating AI" which I found hilarious).The test the article reports about is indicative that for short excerpt, an outdated LLM is enough to convince litterature students that it's OK to read, so while the prose might not be great overall (for long text), it's not eye-gougingly, excruciatingly bad. It's an important step, but not enough to compete -- which is the question the author asks "I can write a book for my own reasons, but I can sell the book only if readers like it more than what they can get from a chatbot". Which is exactly the conclusion: humans need to do better than AI if they expect to sell things. Same with all technologies (we replaced lot of line of works once we discovered we had technologies that does the same thing without needing to pay someone to convince him to do it for us -- Denmark recently decided to close its postal service for letters since people are massively preferring to send instantaneous e-mail rather than paying a courier to move a physical piece of paper, and I didn't see outrage about email stealing our jobs). Creative writers will probably fare better than technical writer (I am pretty sure a LLM could write instructions to build an Ikea piece of furniture right now). Also, there is a market for specific artworks like classical opera with a distinct target from Netflix. There might be a market for artisanal books, but I am not certain that there might be a market for artisanal technical books.</blockquote>

Verification