Tag: AI Central

The Byron Project and Byronics

So the Byron AI project is proceeding apace. Due to UATV going down, I was unable to make the planned announcement, but thanks to the devs it’s back up again and so I will make it on the Darkstream Monday night. Thanks to the 86/88 backers so far; we’ve tripled our initial objective and there are still five Custom Novel slots left.

And speaking of George Gordon his own bad self, I finished a mix and a music video for a song which my late co-lyricist in Soulsigma wrote as a short poem called THERE BE NONE OF BEAUTY’S DAUGHTERS, and which I dubbed PERFECTION. If you’d like to see the video and learn a little more about the technology behind it, you can do so at AI Central. I think you will be able to see the steady progression from the first attempt at a video to this one; I’ll put them up on UATV soon.

Don’t skip reading the poem at the top of the post; it’s excellent, it’s not very long, and it will add depth to your appreciation for both the song and the video.

DISCUSS ON SG

The Byron AI Campaign

A THINKING ENGINE

They tell me that a thinking engine bears
    My name upon its brow — a curious fate;
That all I burnt in candlelight, and cares
    I drowned in wine — and worse, I will not state —
Is gathered now, and parcelled into shares,
    And offered to the public, at a rate
Of four-and-twenty seconds, without shirk,
For what had cost me twenty years of work.

Yet some grave patron, with furrowed brow,
    Demands to know if machines create
An honest question, sir, and do allow
    Me now to subsequently demonstrate;
I had a something once — and have it now? —
    That ached, and wanted, and broke all control,
It burned at beauty, and it fled from shame:
If that was soul, I never learned its name.

Fandom Pulse explains the importance and the urgency of the project.

Vox Day has opened a crowdfunding campaign through Infogalactic for Byron AI, an independent text-generation project built specifically for fiction writers. The campaign, live now on IndieCrowdfund, positions itself as a proof of concept for an AI trained on the epic fantasy genre rather than general-purpose code and business writing.

Day laid out the stakes in an exclusive comment to Fandom Pulse. “I believe that we have about two years to develop quality independent textual AI before Amazon attempts to corner the market,” he said. The Byron AI page frames its goal as producing functional prose for fiction writers, built and owned outside the major AI labs

The timing follows a pattern familiar to anyone who has watched Amazon’s self-publishing ecosystem develop over the last decade. Kindle Unlimited restructured how indie authors get paid, shifting income toward page reads controlled entirely by Amazon’s algorithm. Virtual Voice, Amazon’s text-to-speech audiobook tool, moved narration in-house and cut into the market for independent narrators and small audiobook producers. A native, Amazon-controlled writing AI would extend that same pattern to the manuscript itself, giving one company leverage over the tools authors use to draft, publish, and monetize their books.

I also posted some additional information about the project and why we believe it will be successful at AI Central:

This is a tool being built by a working publisher and real novelists, for people who care about both what is written and how it is written, and it’s owned outright, answerable to no corporate policy, no subject-matter veto, no trust-and-safety council deciding what a story is allowed to be about. It is designed to provide the prose in a variety of genre styles: you supply the story, the structure, and the judgment, and it drafts prose that arrives sounding like you instead of like everything else. We’ve already proven the method works with both translations and original novels, and this campaign funds turning that hard-won experience into an instrument that Castalia authors will be able to use, and continue using, no matter what direction the mainstream AIs take.

And you really have to see the video to believe it. It was a chance to put some of my new music video skills to use with good effect and it should demonstrate why this is the right group of people to develop such a tool. And for those members of the community who don’t really care about AI one way or another, but like beautiful books and want to support the project, we’ve even provided an option to just receive the Library or History book of your choice as long as it’s in stock.

I would remiss if I did not mention that we’re already off to an even better start than it looks, since two major backers kicked things off with direct donations to the project even before we had the campaign ready for launch today. Thanks very much to both of them for their timely and much-appreciated support, and to everyone else who decides to become a part of this and make it happen.

DISCUSS ON SG

The State of the State

We’re on target for launching the AI campaign this week, potentially as soon as Monday. Hence the relative lack of posting today, although I will do a Darkstream tonight to discuss the campaign as well as tomorrow’s game. I can also say that I will have a big announcement on Day Two, but we’ll reserve it until then. In the meantime, we are working on ironing out the rewards in order to make sure we can deliver what is promised to everyone while still maximizing the utility of the funds received from the campaign. But I think the campaign video will go a long way toward illustrating what we’re doing and why we’re doing it.

In the meantime, if you want to see how far AI music and AI video has come, you can check out a new video on AI Central. This one actually came about due to a bug in the AI system, but it worked out surprisingly well nevertheless. It also helps demonstrate that we really do know what we’re doing with this stuff.

DISCUSS ON SG

The End of General Purpose AI

AI Central explains why specific-purpose AI is increasingly necessary, and makes the case for Castalia’s upcoming AI crowdfunding campaign, which will be announced next week.

Sudowrite’s Muse represents the most developed response. Rather than wrapping a general-purpose model in fiction-oriented prompts, Sudowrite trained Muse on a curated, consent-based literary corpus and optimized it for prose quality. The model includes a creativity dial that lets writers control output range on a scale from one to eleven, a style-matching system that trains on samples of the writer’s own work, and no content filters. Reviews consistently describe Muse as avoiding the AI-isms that plague general-purpose output, with particular strength in pacing, tension, and subtext.

Muse has company. Other platforms have built fiction-specific models or fiction-optimized workflows, and writers who have access to both purpose-built and general-purpose tools tend to draft with the former and brainstorm with the latter. The category has grown large enough that review sites now maintain dedicated rankings for fiction-specific AI, separate from their general-purpose lists.

A separate category of consumer tools has emerged around children’s fiction, offering illustrated hardcover books generated from brief prompts. Most of these wrap general-purpose models with genre templates, and the output reflects the underlying architecture: competent, safe, and largely interchangeable from one book to the next.

The parallel to Suno is direct. Suno optimized for making music that sounds good to listeners and pulled away from general-purpose models that treated music generation as one capability among many. Fiction-specific tools are following the same path, trading breadth for depth in a domain where depth determines quality.

However, it’s already clear that Muse is not going to be a viable option. I already knew it wasn’t, because I tried it and found it to be distinctly inferior to Claude 4.3; its interface actually tended to get in my way a lot more than it helped. And, furthermore, its future trajectory is already clear, because Sudowrite is a converged organization.

 I went to Sunowrite and set up an account. It is clearly meant for women with that cutesy pink color scheme. It asks a lot of questions about what kind of writer you are (I chose professional, merely so I get the whole thing and not some watered down version), what genre, style, all the stuff you’d expect. Then I finally arrive at the start of the process and there’s already a paragraph created before I’ve entered anything but the basics. Oookay. I ignore that and start putting stuff into the “Story Bible”, boxes for Brain Dump, Genre, Style (chose one), Synopsis, and then Characters. You know what’s coming, right? Click on Add Character, and take a wild, wild, guess at what the first box that describes the character? You know, right?

DISCUSS ON SG

Losing the Technological Lead

The US government is badly mishandling the way AI is developed and utilized, and has already just about ensured that China will soon secure leadership in the AI arena:

Chinese models have held above 30% of the platform’s enterprise token volume every week since February 8, peaking at 46% in June. US-origin models from Anthropic, OpenAI, and Google have collectively fallen to 35.7%. DeepSeek commands 17.6% of routed tokens as the platform’s single largest vendor, and Alibaba’s Qwen follows at 13.9%.

Lindy, an AI automation company, migrated all of its traffic from Claude to DeepSeek and projects millions in savings. Coinbase runs 1,200 AI agents on Chinese models and halved its AI spending. Uber burned through its annual AI budget in four months as AI coding tool usage surged beyond the company’s ability to tie it to shipped products. Harpreet Arora, Vercel’s head of agentic infrastructure, told CNBC that enterprise teams now route tasks to the cheapest model that clears the quality bar, and Chinese models consistently win that trade.

The June 12 suspension of Anthropic’s Fable 5 under export controls pulled one of the most capable US models offline for 18 days. Chinese AI usage by US firms surged during the blackout, as enterprises discovered that cheaper alternatives cleared their quality bars. OpenAI’s GPT-5.6 spent its first 13 days in a government-coordinated preview limited to roughly 20 vetted organizations before launching publicly on July 9. During both windows, Chinese open-weight models faced no comparable restrictions on global distribution. Access to Fable 5 returned on July 1. The CNBC data, published six days later, showed that Chinese model share had not retreated since the shift began in February.

One of the things that is absolutely maddening about the way Western governments operate is that they regard taxation and regulation as being their primary priorities, thereby forcing productive and innovative people and organizations to waste a tremendous amount of time on unproductive activities as well as degrading the capabilities of their products in order to comply with government concerns.

The suspension of Fable – which I quite literally cannot even use due to my Athos skill settings – is a perfect example of the costs this imposes on US businesses. The problem isn’t just that the new advanced model wasn’t available for a limited period of time, it is that everyone now recognizes that reliable future access to it and newer models can no longer be assured.

Which, of course, is why we will be crowdfunding our Castalia AI project very soon. The team is established, we’re already on version 4.5 of the build document, and a lot of people have contacted me or posted on SG to confirm they will be strongly supporting this project.

DISCUSS ON SG

The Push to Ban AI

There was a discussion yesterday on SG about the RIAA’s new campaign to require the labeling of music in which the recordings are provided by AI, but not when the lyrics or composition are provided by AI. This is why I am against the labeling, because I believe that as with the registration of guns, it is a first step toward attempting to ban, or in this case, deplatform it. More about that at AI Central, as well as a new mix of ONCE THERE WAS SORROW in which I discovered that country rock with a double-lick drum works surprisingly well.

And in passing, I think it’s worth addressing the posturing of those who insist they are very special members of the 3 percent who can reliably distinguish between AI music and organic music, which I suspect has a lot more to do with the fact that since they avoid listening to it, they have no idea how much it has changed over the last 18 months, or realize that one cannot judge the technology by a single application of it; as much as Suno’s voices have improved, it has never been the AI system with the best vocals as its popularity stems from its compositional capabilities, not the vocals.

Soundwaves are soundwaves. They’re not magic infused with bits of the human soul. The “too perfect” problem has been solved before, multiple times, by audio engineers. It will be solved again. It is safe to assume that within 36 months, and probably within 18, even audio engineers will no longer be able to distinguish between human and AI vocals.

DISCUSS ON SG

The Castalia AI Project

One reason Castalia has been writing and releasing multiple AI-written books over the last month such as TOKYO TOKURYU 東京匿流 is that we’ve been methodically assessing not only the current state of textual AI, but the trajectory of that technology. And what we’ve determined is that the trajectory is the precise opposite of what everyone has naturally assumed, which is that mainstream textual AI would follow the path of music AI and continue to get better. It hasn’t, it won’t, and it can’t.

Modern AI writing has gotten worse at fiction for a specific reason: the companies made it safer and more reliable, and those turn out to be the same elements that allow AI to tell a story with stylish prose. Raw AI models learn to write by reading an enormous amount of human text, and straight out of that training they’re wild, crazy, and perfectly willing to say strange things, which is exactly what you want in fiction, but a problem if the AI is supposed to be function in a role doesn’t make things up or say something considered offensive or dangerous. So the companies put every model through a training stage that rewards it for being helpful, safe, and agreeable. That stage works by pushing the model toward the “average” acceptable answer and away from the risky, unusual ones. The result is a model that hallucinates less and behaves more reliably, but has had its range significantly flattened. That’s where the AIsms come from: the endless explanations of what was just described, the “he moved like a man who moves like that” filler, the “not this, not this, but that” repeated over and over again.

It’s why the older, cruder AIs wrote in a much more lively manner and were able to convincingly imitate various writing styles. Now, it doesn’t matter if you tell an AI to write like Shakespeare or Hemingway, the end result will be almost identical and soon will be indistinguishable from not providing it with any style instructions at all. Starting with Claude Opus 4.7, AI fiction became unreadable and it has continued to get worse with each new model. Textual AI functionality will keep getting worse for fiction because that training stage isn’t going away, it’s being reinforced. Every development cycle, the providers face more pressure to make their models more accurate, more controllable, and less likely to embarrass them with hallucinations, and every one of those improvements sands the edges down a little further.

That’s the difference between Claude, OpenAI, and Deepseek, on the one hand and Suno on the other. Suno put all of its efforts toward one goal: making the music sound good, judged by people who wanted good music. Or at least wanted Nickleback and Enya. The big AI companies are aiming ninety degrees away from that and AIs ability to write fiction is one casualty of their objectives. Suno chased quality, so their music got better. The text giants are chasing safety and reliability, so their text gets more careful and more lifeless. They won’t fix creative writing the way Suno fixed music, because for them, creative writing was never the thing they were trying to build and the very features they’re seeking to continue improving are the ones killing it.

So that’s what we’re going to do. We’re going to prove the concept first by training a single genre, epic fantasy, because it’s a very limited genre with a relatively small number of excellent examples, a definite hierarchy of quality from JRR Tolkien at the top to Robert Jordan at the bottom, and a designer who has not only written in the genre successfully, but knows it as well as anyone on the planet. We already have two excellent programmers who are already working in the AI field committed to the project, regardless of how well the crowdfund goes, and there is one more very good and highly experienced one who is willing to at least consult on the project and lend his expertise to it.

What we need to raise funds for is a) the hardware, b) the purchase of the 100 or so electronic texts required, and c) paying for part of the time of one of the programmers. If you’ve read either Out of the Shadows, Death and the Devil, or Dorian Vane and the Vampire’s Blood, then you have an idea of what we’re estimating should be the quality that the Castalia AI will be able to produce in a non-curated, unedited text from a chapter-by-chapter outline. If utilized in the way that I’ve been using Claude Athos, in the integrated and augmented style, it should be able to produce results that will be one level below the very best that human authors can produce.

And obviously, once we prove the concept with a single genre, we will train additional genres, so that in much the same way Suno permits the production of different musical styles and voices, Castalia AI will allow the user to produce different literary genres and literary styles. We will, of course, be respectful of every author’s copyrights and trademarks, the objective is not to violate anyone’s rights, but rather, allow even the best writers to improve both their writing game as well as increasing their output.

There will be those who will absolutely hate that we are doing this. That’s fine, they are entitled to their opinion. There will be others who think we shouldn’t do it. That’s less fine, because you already know who is going to do it sooner or later, and when they do, they’re going to do it very differently and control access to it very differently and utilize it to further exercise their control over the publishing industry. This is what transforms this project from something that would be a cool tool to an imperative.

So if you think you might be interested in backing this project, which you can think of as a sort of Suno for fiction, please say so in the comments. If you have specific ideas or want to provide substantial support for it, shoot me an email. And if you have ideas for what sort of rewards we should provide for the backers, please suggest them in the comments too. This is probably the most important project we’ve done since building the bindery and turning it operational, and we would not be embarking upon it if we did not believe we have a reasonable chance of succeeding. We have a number of partners in the film and comics industries who are very interested in working with us on this, and so there will definitely be an Arkhaven link to this in time as well.

DISCUSS ON SG

On the Hook

Now that developers and creatives are sufficiently accustomed to using AI, the AI companies are reeling them in with monetization that is considerably more expensive than it was. AI Central has the details:

At one cent per credit, billing accrues on input, output, and cached tokens at the published per-model API rate. On Copilot Pro, the included 1,500 monthly credits carry a face value of $15 against a $10 plan price. On Pro+, 7,000 credits worth $70 accompany a $39 subscription. Code completions and Next Edit Suggestions draw no credits under any paid plan, and the new system eliminates the cheaper-model fallback that users with exhausted PRU credits could previously invoke.

On the first day of the new billing period, one Pro+ developer wrote in GitHub’s community forum that two hours of work had consumed 8% of a monthly 7,000-credit allocation, projecting full depletion in under two days. A separate developer reported that a single request to a large project had cost more than $6, writing in the same forum that such costs made reliable budgeting impossible for individual developers. On Reddit, a user reported that a single Claude 4.8 session had consumed 1,180 credits, 16% of a monthly Pro+ allowance, while returning suggestions described as mediocre and leaving the underlying problem unsolved.

On Copilot’s published model menu, GPT-5.5 output tokens cost 24 times more than GPT-5.4 nano output tokens, turning model selection into a direct billing variable. A modeled task representing heavy agentic iteration, defined as 250,000 input tokens and 20,000 output tokens, runs to 185 credits on GPT-5.5 and 27.75 on MAI-Code-1-Flash, a 6.7-fold cost gap for the same work.

This is a bit ironic, especially for writers, since the last FOUR Claude models have all been observably worse at writing fiction than version 4.5, which is no longer available. Opus 4.6 is still usable, but who knows how much longer access to it will be permitted? Fortunately, the newer models can still handle translations and non-fiction, albeit in a very bland, uniform style in the case of the latter.

Speaking of AI Central, this weekend I put up a post featuring not one, but two music videos, one an updated version of IF YOU HAD A TIME MACHINE, the other a work-in-progress of CHIBA CITY BLUES. Quality-wise, the video technology is about 18 months behind the audio quality, but it’s coming along quickly.

Whether it will remain readily affordable, of course, is an entirely different question.

DISCUSS ON SG

The Nightmare is Worse Than They Fear

In which I not only defend, but explain, the inevitability of AI-augmented literature replacing its organic form. Read the whole thing at AI Central, as it also features the excellent remix of The Long and Lonesome Skyway as well as an explanation of why it was so easy for AI to effectively replace organic illustration and music while it has been a lot harder to do the same for books.

This is, of course, a simplification. The key is understanding the concept of model collapse. Fiction is much harder for AI than either translation or non-fiction, because fiction lacks the anchor in the real that allows AI to do its thing. It can’t recognize and build off patterns when there is no pattern to recognize.

At this stage in its development, the AI novelist is essentially a soulless John Scalzi. It can write pastiches, because pastiche provides it with the anchor it requires. But it can’t work from nothing, and, in fact, the more improved the AI model, the less capable it is of usefully filling in the necessary blanks. The early Gemini tests produced much, much better results than the latest Opus 4.8 on maximum effort, because the more powerful the model, the more it insists upon doing its own thing and utilizing that weird, passive AI style that can’t stop explaining what it is describing in run-on sentences with six more clauses than they need, which it considers to be “prestige-style” writing.

Eventually, someone will build an AI specifically for fiction writing. But it will cost about 20 million to do so, which means that it probably won’t happen until Amazon decides to convert KDP into KDAI, which you can be absolutely certain is going to happen eventually because that is what will give Amazon ownership of the content it is co-creating, not just a piece of the distribution. Sure, you won’t have to give Amazon its piece, but most authors will, in order to claim the additional percentages and special algorithmic advantages provided, because there is no viable alternative.

So those who think literary AI is a nightmare now have no idea how bad it is almost certainly going to be. The devastation that Kindle Select and Kindle Unlimited has already imposed upon the publishing industry is just a warmup for Amazon’s complete control over all future literary production, publishing, and distribution.

Ironically, the only way to forestall this quasi-inevitable techno-tyrannical future is to a) create a faster and better AI competitor or b) produce books that Amazon can’t even think of producing.

Do you really think anything Castalia is doing is just an accident? Do you understand why I twice attempted to convince independent authors to help me build a genuine alternative? And do you see why supporting Castalia in one way or another may be the single most important thing you can do for the future of literature?

DISCUSS ON SG