I'm prepared to pledge $1,000 for the community access tier, with one question first: does that tier include voting on genre/style selection, or is that exclusive to the lower tier?
If voting isn't currently included at $1,000, would you be willing to extend it to backers at that level? The style selection is the part I most want a say in, so as it stands I'd pledge at the voting tier instead. If the $1,000 tier carries both, I'm in at $1,000.
I gladly pledged 1000 as I've recently gotten into writing. The current AI writing tools are decent but run into the same problems listed in this article. Byron AI sounds like exactly what I need. I'm great at world building but suck at Prose. Looking forward to the release.
What technical approach(es) are you looking to explore? Fine-tuning an open source chinese model? RL might work but also could handcuff the creativity...I don't really know. Curious too if you see this more as an agent w/ pluggable backends and possibly subagents (something like Grok Heavy), or more of a tokens_in -> tokens_out layer, that others can assemble into more complex agentic patterns.
Also quite curious how you think about evals / measurement of the success, and the QA process generally for a product like this. Very hard to write a spec that could be verified deterministically that it has sufficient creativity. Although, maybe claude/gpt/grok actually would be good judges of this, even if they can't summon the creativity within themselves...
In this post you said "one factor that will be very important is deciding which of the training works to overweight, or in our terminology, to assign the gold standard."
and then
"I’d be interested in hearing is your suggestions on what the top ten epic fantasy books would be. Not series, books. Remember, the focus here is on writing style, not worldbuilding, not plotting, and not characters per se. So the limits are one author, one book."
After reading that I was under the impression that there is a specific set of books you want to use for training the AI; my thought was at level $XX add another book to this list.
We have a core list of 100 different books from different authors in three different weightings. What you're suggesting is a good idea, but for a different genre. There just isn't enough epic fantasy to include any more books beside more from the same author, which is not desirable.
It is a sorrow to not have 2,500 lying about for just such an occasion, definitely have far more ideas than capabilities to realise them, and having Vox helm an one of this teller's ideas would be fascinating. Doubtless either Jon or Vox will do good work with whatever idea they have to deal with.
50 can be done. Depending on circumstances, might be able to go for the hardcover at 150, got 29 days to gather loose funds...
Not completely destitute, but most available funds aren't truly available until this teller has fulfilled his own campaign obligations. Fixed liquid income is certainly convenient, the cost of one's actions oh well. Should be in much better standing next year, which is, unfortunately too late for this important campaign.
Presumably the finished product distantly down the line will be a subscription service?
Incidentally you probably have already thought of this, but if you can make an end to end idea to publication pipeline integrated into it with differently priced options from ebook to leather...
As of yet nobody has truly capitalised on the fact that most people do not want to be their own publisher and advertising group, a one man band. The larger majority of people in the field are distressingly timid, quite the opposite of the more prominent ones.
Admittedly it may be that there's not a lot of money in it, but this teller's guts tell him that fulfillment is something even non AI users would be attracted to.
Harness that griffon, and may this go very well!
EDIT: Just saw the 1,000 option, community access people! 75,000 words is just fine! Now this teller deeply regrets not having affairs in order and merchandise on sale. Remember folks, if you don't have your house in order your capacity to support is greatly limited, don't be like this fellow!
Is Byron intended to remain an in-house instrument for Castalia authors, or is there a plan to make it available for purchase to the general consumer market once the proof of concept is validated?
There is no "business plan" as yet. We're entirely focused on making it work better than general AI. If it works well enough to be a community or commercial service, we'll figure out how to make it available, starting with the backers. The "in-house instrument" is more of testing phase; writers like JDA and me don't need as many bells and whistles as most users would.
I'm prepared to pledge $1,000 for the community access tier, with one question first: does that tier include voting on genre/style selection, or is that exclusive to the lower tier?
If voting isn't currently included at $1,000, would you be willing to extend it to backers at that level? The style selection is the part I most want a say in, so as it stands I'd pledge at the voting tier instead. If the $1,000 tier carries both, I'm in at $1,000.
I gladly pledged 1000 as I've recently gotten into writing. The current AI writing tools are decent but run into the same problems listed in this article. Byron AI sounds like exactly what I need. I'm great at world building but suck at Prose. Looking forward to the release.
Thank you. If it's possible to eliminate them, we will do so.
$2500 tier is a sweet deal
I need to check the budget hard — the leather book would be lovely and access to the AI more so but it will depend on cashflow
What technical approach(es) are you looking to explore? Fine-tuning an open source chinese model? RL might work but also could handcuff the creativity...I don't really know. Curious too if you see this more as an agent w/ pluggable backends and possibly subagents (something like Grok Heavy), or more of a tokens_in -> tokens_out layer, that others can assemble into more complex agentic patterns.
Also quite curious how you think about evals / measurement of the success, and the QA process generally for a product like this. Very hard to write a spec that could be verified deterministically that it has sufficient creativity. Although, maybe claude/gpt/grok actually would be good judges of this, even if they can't summon the creativity within themselves...
Is it in scope to request the "Custom Novel - VD" reward to be a story other than _A Grave of Gods_, but still set in Selenoth?
Certainly.
Backed!
Just backed. What Castalia has already done is both impressive and extremely important. This will be too.
Can you set a level at 50 or 100 bucks so the backers at that level can choose an additional training text?
Do you mean give them a vote on what the next genre or literary style will be? Please be a little more specific.
But sure, we can add another tier. And we already have a $50 tier.
https://voxday.net/2026/07/11/top-10-epic-fantasies/
In this post you said "one factor that will be very important is deciding which of the training works to overweight, or in our terminology, to assign the gold standard."
and then
"I’d be interested in hearing is your suggestions on what the top ten epic fantasy books would be. Not series, books. Remember, the focus here is on writing style, not worldbuilding, not plotting, and not characters per se. So the limits are one author, one book."
After reading that I was under the impression that there is a specific set of books you want to use for training the AI; my thought was at level $XX add another book to this list.
We have a core list of 100 different books from different authors in three different weightings. What you're suggesting is a good idea, but for a different genre. There just isn't enough epic fantasy to include any more books beside more from the same author, which is not desirable.
I like this idea a lot.
TAKE MY MONEY
It is a sorrow to not have 2,500 lying about for just such an occasion, definitely have far more ideas than capabilities to realise them, and having Vox helm an one of this teller's ideas would be fascinating. Doubtless either Jon or Vox will do good work with whatever idea they have to deal with.
50 can be done. Depending on circumstances, might be able to go for the hardcover at 150, got 29 days to gather loose funds...
Not completely destitute, but most available funds aren't truly available until this teller has fulfilled his own campaign obligations. Fixed liquid income is certainly convenient, the cost of one's actions oh well. Should be in much better standing next year, which is, unfortunately too late for this important campaign.
Presumably the finished product distantly down the line will be a subscription service?
Incidentally you probably have already thought of this, but if you can make an end to end idea to publication pipeline integrated into it with differently priced options from ebook to leather...
As of yet nobody has truly capitalised on the fact that most people do not want to be their own publisher and advertising group, a one man band. The larger majority of people in the field are distressingly timid, quite the opposite of the more prominent ones.
Admittedly it may be that there's not a lot of money in it, but this teller's guts tell him that fulfillment is something even non AI users would be attracted to.
Harness that griffon, and may this go very well!
EDIT: Just saw the 1,000 option, community access people! 75,000 words is just fine! Now this teller deeply regrets not having affairs in order and merchandise on sale. Remember folks, if you don't have your house in order your capacity to support is greatly limited, don't be like this fellow!
Only $10,000 goal? AI development must be much further along than I thought. The perks look great.
I'm in, because I want it, too.
Is Byron intended to remain an in-house instrument for Castalia authors, or is there a plan to make it available for purchase to the general consumer market once the proof of concept is validated?
There is no "business plan" as yet. We're entirely focused on making it work better than general AI. If it works well enough to be a community or commercial service, we'll figure out how to make it available, starting with the backers. The "in-house instrument" is more of testing phase; writers like JDA and me don't need as many bells and whistles as most users would.
And if so, what is the expected price for a personal copy?
It's not a product, more of a service. We'll price it reasonably if there is demand for it. But right now, it's not possible to say.