Using LLMs to brainstorm new sources of income
99 views
Skip to first unread message
Tom Gally
May 11, 2025, 9:34:35 AMMay 11
to hon...@googlegroups.com, swe...@groups.io
Here’s a report about a little project I spent a couple of hours on today.
It was spurred by an article in this morning’s Nikkei about the increasing popularity of self-published fiction in Japan, with the news hook being the 文学フリマ held in Tokyo today:
For the past couple of years, I have speculated—rather idly—that there might be large untapped demand for the translation into other languages of Japanese self-published fiction and that recent LLMs might be able to help meet that demand. But I had thought about it only from the viewpoint of people who might want to read the fiction. This article made me realize that there might also be writers who want their works translated but can’t afford to pay human translators.
After looking at the 文学フリマ website, I composed the following prompt and fed it to Gemini, ChatGPT, and Claude in their research modes:
Here are the results:
I learned a lot from Gemini’s and ChatGPT’s reports, though Claude’s is not useless. Note that I had ChatGPT check the references in the Gemini report and that they seem largely correct. Earlier versions of Gemini Deep Research frequently hallucinated references.
After reading the Nikkei article, I wondered if there might be a potential business opportunity for someone experienced with both translation and LLMs to help writers get their works translated into other languages. The quality would not be as good as by skilled human translators, of course, but it should be better than what the writers could get by doing the machine translations themselves.
So I asked ChatGPT o3 to prepare a business plan. Here it is:
I myself am not interested in starting such a business. Rather, I offer this story to freelancers as an example of how one can use the latest reasoning models with research capability to deepen one’s knowledge and understanding about a field and to assess potential new sources of income.
Tom Gally
Yokohama, Japan
It was spurred by an article in this morning’s Nikkei about the increasing popularity of self-published fiction in Japan, with the news hook being the 文学フリマ held in Tokyo today:
https://www.nikkei.com/article/DGXZQOUC10E1Y0Q5A410C2000000/
For the past couple of years, I have speculated—rather idly—that there might be large untapped demand for the translation into other languages of Japanese self-published fiction and that recent LLMs might be able to help meet that demand. But I had thought about it only from the viewpoint of people who might want to read the fiction. This article made me realize that there might also be writers who want their works translated but can’t afford to pay human translators.
After looking at the 文学フリマ website, I composed the following prompt and fed it to Gemini, ChatGPT, and Claude in their research modes:
I am impressed by the amount of self-published literature that is written and shared in Japan, such as through the 文学フリマ https://bunfree.net/ . I wonder if there have been any moves to help those authors get their work translated into English and other languages, as recent advances in LLM-based machine translation would seem to make that possible. Please find out whatever you can about the translation of self-published fiction and other forms of literature, both serious and popular. I am especially interested in noncommercial efforts by the authors themselves or by people helping the authors. Focus primarily on literature written in Japanese, but also mention similar efforts (if any) in other languages as well.
Here are the results:
https://docs.google.com/document/d/1dVpqHqSKV3UdYKJwNY5Ufb7O0yIMopja1pEKf2mIS_s/edit?usp=sharing
https://chatgpt.com/share/68209307-a8fc-8011-91a2-aeac49360171
https://claude.ai/public/artifacts/54ae18f0-11a1-441a-a9ab-3ac088153a28
I learned a lot from Gemini’s and ChatGPT’s reports, though Claude’s is not useless. Note that I had ChatGPT check the references in the Gemini report and that they seem largely correct. Earlier versions of Gemini Deep Research frequently hallucinated references.
After reading the Nikkei article, I wondered if there might be a potential business opportunity for someone experienced with both translation and LLMs to help writers get their works translated into other languages. The quality would not be as good as by skilled human translators, of course, but it should be better than what the writers could get by doing the machine translations themselves.
So I asked ChatGPT o3 to prepare a business plan. Here it is:
https://chatgpt.com/share/6820a17f-5404-8011-953a-768cdccf05e3
From the conclusion: “This business plan demonstrates that a one-person operation leveraging state-of-the-art LLMs can viably offer fast, low-cost literary translations to a niche of budget-conscious Japanese authors.”
Tom Gally
Yokohama, Japan
Fred Uleman
May 11, 2025, 7:12:19 PMMay 11
to SWE...@groups.io, hon...@googlegroups.com
Tom,
Thanks. Interesting.
Just read the first part of the business plan, but:
Was thinking about this last night before going to bed and:
Were I interested in actually doing something (I am not), I would position it not as a translation service but as a literary agent. There are all these people writing for very small audiences -- some for just themselves -- but they would like to be more widely read. So be a literary agent and offer to represent them. Lock them in.
And to represent them, use the free machines to do the translations of snippets and put the snippets on your website as "waiting to be discovered talent." Then send come-on blurbs and links to all the publishers AI can find. Ask them to take a look and see what they like. Because you represent these people and can make it happen.
Think of it as a consolidated yard sale for writers where you provide the yard and take a cut.
Sure, you might want to add a disclaimer to the site that all of these "translations" were machine-generated and are just to give the publishers an idea of what is there --- that the actual translations will be better --- but position yourself as the next step beyond/above the translation service idea.
fwiw.
- -- --- ---- --- -- -
Fred "I used to be a translator" Uleman
_._,_._,_
Groups.io Links:You receive all messages sent to this group.
View/Reply Online (#1289) | Reply to Group | Reply to Sender | Mute This Topic | New Topic
Reading this as email? You can view earlier messages online at https://groups.io/g/SWET-L/topics.
Your Subscription | Contact Group Owner | Unsubscribe [ful...@gmail.com]
_._,_._,_
Tom Gally
May 12, 2025, 9:29:57 AMMay 12
to hon...@googlegroups.com, swe...@groups.io
Rereading the business plan that ChatGPT o3 produced for me yesterday, I am struck both by its seemingly ethical stance—presenting the proposed services honestly and without exaggeration—and by its apparent insights into some key aspects of the translation process. Neither of those was present in my prompt to it.
Some examples, in no particular order, emphasis added:
Something it missed, and that I had forgotten about, too, is that the LLMs mentioned—ChatGPT, Claude, Gemini, etc.—might refuse to translate erotic passages, which are not unusual in self-published fiction. There are other topics that appear in literature but would probably trip the safety switches of commercial LLMs, such as politics, racism, and violence. If someone launches this business, they will need to be attentive to such issues.By using advanced AI prompts and reviewing outputs from multiple LLMs (e.g. GPT-4o, Claude 3.5, Gemini 1.5), the business can deliver acceptable literary translations quickly and cheaply.
The result is a reasonably high-quality translation without the premium price tag.
The growing acceptance of AI-assisted content suggests the market for such hybrid services will expand, especially if the quality approaches human level.
The human translator provides oversight: ensuring names, tone, and nuances are correctly conveyed and doing light editing for natural English.
Thus, there is a clear need for an affordable “machine+human” translation service that handles the process end-to-end, maintaining context and fluency.
Effective prompts are crucial – for example, instructing the LLM to maintain the narrative voice or to output in natural novel-like English paragraphs.
The human can quickly fix any mismatches (e.g. if an honorific was dropped or a name was translated differently in various chapters).
Light fluency editing: making minor tweaks so the final English text reads smoothly and retains the author’s style.
The finished translation is sent to the client, typically accompanied by a brief note on any sections of uncertainty or suggestions (if any cultural nuance needed explanation).
In the future, building a repository of common terms or stylistic choices could be useful (especially if authors come with series or sequels). At launch, this isn’t critical; the LLMs handle context internally, and each project is separate. No expensive CAT software is required.
Pure quality and literary nuance from an experienced human translator is still higher. A human translator can localize idioms, choose perfect diction, and adapt humor or poetry in ways AI cannot fully achieve yet. We must be honest that our service provides “good enough” translation, not award-winning prose. In competitive positioning, however, this is acceptable because our clients are not expecting perfection at this price – they want a serviceable English version to share with fans or test the market.
Nevertheless, I remain impressed at how thorough and well-organized a business plan this model could produce from my hastily dictated prompt. It reinforces my feeling that the key to competing with AI will be to emphasize the value, visibility, and identity of us as individuals. If I were launching this business myself, I would attend all those 文学フリマ and other events in person, talk to as many authors as I could, and put a photo of myself on all of my publicity materials.
Tom Gally
Bill Lise
May 12, 2025, 9:56:00 AMMay 12
to hon...@googlegroups.com
>>It reinforces my feeling that the key to competing with AI will be to emphasize the value, visibility, and identity of us as individuals. If I were launching this business myself, I would attend all those 文学フリマ and other events in person, talk to as many authors as I could, and put a photo of myself on all of my publicity materials.<<
Yes, professionals need to emphasize their value as individuals, and it’s probably easier for freelancers to do that with other individuals than with prospective corporate clients, which is what will be left shortly.
I see a parallel to this idea in things that have been voiced by translators on LinkedIn. Several of them have said that, as their agency clients have left them, they have seen an increase in individual clients. The other factor in this client demographic shift is that it very difficult for a freelancer to develop direct corporate clients.
I assume these authors are individuals, and while they probably cannot spend much money on translation, given the likely lack of hope for any substantial return, if the cost of producing the “good enough” literary translations is low enough, this kind of business might work.
One question I have is the reference to not needing “expensive CAT software.” I wonder if this assumes that the AI usable for such work will be essentially free?
Bill Lise*********************************************
リゼ ウィリアム
Bill Lise
*********************************************
--
You received this message because you are subscribed to the Google Groups "Honyaku E<>J translation list" group.
To unsubscribe from this group and stop receiving emails from it, send an email to honyaku+u...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/honyaku/CACnS3NcmXRiv36bs%3Dc7XvM9JR56-FFWJB2GOQFSkdqOywswsVw%40mail.gmail.com.
Oroszlany Balazs
May 12, 2025, 10:13:47 AMMay 12
to hon...@googlegroups.com, swe...@groups.io
Tom,
How often do you buy self-pub books in English or Japanese? What percentage of these self-pub books are translations?
Putting aside the question of quality (let's assume it is good enough - although "good enough is a red flag for me."(T)hey want a serviceable English version" - this is the standard of a software startup, not a serious literary translator.), and not considering the copyright issues (is it WFH? Does the translator/prompter create copyright?), it is easy enough to develop a pipeline for _content creation_. But would this new entity be helpful to the original authors? I doubt so - even with low translation cost, most books will never return the investments. Is it good for the translator? I don't think so - having dozens of books under your name in only a few years is not always the best recommendation if you want to transfer your skills to more traditional authors. Is it good for the readers? Well, with the abundance of human-translated (Japanese) literature already available in English (often at great sale prices) and fan translations, I would be really surprised if there is a significant unmet need for these books to be translated.
It sounds like a resource-heavy solution for a problem that doesn't really exist.
Best regards,
Balazs Oroszlany
How often do you buy self-pub books in English or Japanese? What percentage of these self-pub books are translations?
Putting aside the question of quality (let's assume it is good enough - although "good enough is a red flag for me."(T)hey want a serviceable English version" - this is the standard of a software startup, not a serious literary translator.), and not considering the copyright issues (is it WFH? Does the translator/prompter create copyright?), it is easy enough to develop a pipeline for _content creation_. But would this new entity be helpful to the original authors? I doubt so - even with low translation cost, most books will never return the investments. Is it good for the translator? I don't think so - having dozens of books under your name in only a few years is not always the best recommendation if you want to transfer your skills to more traditional authors. Is it good for the readers? Well, with the abundance of human-translated (Japanese) literature already available in English (often at great sale prices) and fan translations, I would be really surprised if there is a significant unmet need for these books to be translated.
It sounds like a resource-heavy solution for a problem that doesn't really exist.
Best regards,
Balazs Oroszlany
--
Tom Gally
May 12, 2025, 7:54:54 PMMay 12
to SWE...@groups.io, hon...@googlegroups.com
A few clarifications.
The customers for this hypothetical business would be writers who publish their literary works either electronically or on paper without expectations of earning money on them. While some self-published writers do become successful commercial authors, they are a tiny minority. People who publish their fiction on sites like Pixiv or Storys.jp or who pay to have their books printed for distribution to friends and at places like 文学フリマ do it for the creative satisfaction and for the opportunity it gives them to interact with like-minded people. I suspect that many people reading this have similar creative outlets that bring them satisfaction but no money.
In the case of literature, both popular and serious, the language barrier has kept amateur writers trapped within their linguistic communities. Only well-to-do people are able to pay human translators to translate creative works that do not have a commercial market. When I was freelancing, I did take on a few jobs like that—not novels, but some essays and other works that wealthy amateur writers wanted to release in English. I remember a talk by Fred Uleman at an IJET 20 or 25 years ago in which he described translating a book for one of the richest men in Japan. There’s not much of that work going around, though.
So my idea, which I brainstormed further with the various AIs, was that there might be a lot of self-published writers who would like to share their writing in other languages. Until recently, that was impossible because they couldn’t afford to pay human translators to produce translations that would not earn any money. Even with widely available machine translation now, somebody who is not sophisticated about translation or about the target language will not get very good results. A person sophisticated about translation could design prompts and agentic workflows that would produce passable translations at low cost, and they could use their expertise and experience as a selling point for their services. As the ChatGPT business plan made clear, the service provider would be honest about the limitations on the quality.
The person running this business would not actually line-edit the translations. As I said in the prompt to ChatGPT, the person would spend only one or two hours per book. The translator would just design the workflows and spot-check the results.
If you haven't tried these latest models like ChatGPT o3, Cloud 3.7 Sonnet in thinking mode, or Gemini 2.5 Pro, well, let me tell you, they are not DeepL or Google Translate. You can give them complex tasks and high-level instructions about strategies to follow—including translation and editing workflows—and often they do quite well.
When translating novels using these models, the translator would design prompts to deal with issues that the translator knows can cause problems for machine translation of fiction as well as prompts for the particular work. The translator would then do a quick check to see how well those prompts have been followed, refine the prompts if necessary, and run the translation through the LLM again. These agentic models can also do basic line editing, like checking for omissions and inconsistencies. Based on the tests that I have run, I think they would do okay at first but that some problems would emerge. The results would gradually get better as the prompts and workflow are refined.
My main purpose in posting that report was not to suggest an actual business plan, even though it might be viable. Rather it was to show how powerful these most recent models are getting with higher level thinking. The space of intellectual skills that remain uniquely human continues to shrink. That's why, in my last post yesterday, I reiterated the importance of foregrounding one’s personal identity and engaging in face-to-face interaction. Show your potential customers that you are human and make them want to interact with you instead of with a machine.
Tom Gally
The customers for this hypothetical business would be writers who publish their literary works either electronically or on paper without expectations of earning money on them. While some self-published writers do become successful commercial authors, they are a tiny minority. People who publish their fiction on sites like Pixiv or Storys.jp or who pay to have their books printed for distribution to friends and at places like 文学フリマ do it for the creative satisfaction and for the opportunity it gives them to interact with like-minded people. I suspect that many people reading this have similar creative outlets that bring them satisfaction but no money.
In the case of literature, both popular and serious, the language barrier has kept amateur writers trapped within their linguistic communities. Only well-to-do people are able to pay human translators to translate creative works that do not have a commercial market. When I was freelancing, I did take on a few jobs like that—not novels, but some essays and other works that wealthy amateur writers wanted to release in English. I remember a talk by Fred Uleman at an IJET 20 or 25 years ago in which he described translating a book for one of the richest men in Japan. There’s not much of that work going around, though.
So my idea, which I brainstormed further with the various AIs, was that there might be a lot of self-published writers who would like to share their writing in other languages. Until recently, that was impossible because they couldn’t afford to pay human translators to produce translations that would not earn any money. Even with widely available machine translation now, somebody who is not sophisticated about translation or about the target language will not get very good results. A person sophisticated about translation could design prompts and agentic workflows that would produce passable translations at low cost, and they could use their expertise and experience as a selling point for their services. As the ChatGPT business plan made clear, the service provider would be honest about the limitations on the quality.
The person running this business would not actually line-edit the translations. As I said in the prompt to ChatGPT, the person would spend only one or two hours per book. The translator would just design the workflows and spot-check the results.
If you haven't tried these latest models like ChatGPT o3, Cloud 3.7 Sonnet in thinking mode, or Gemini 2.5 Pro, well, let me tell you, they are not DeepL or Google Translate. You can give them complex tasks and high-level instructions about strategies to follow—including translation and editing workflows—and often they do quite well.
When translating novels using these models, the translator would design prompts to deal with issues that the translator knows can cause problems for machine translation of fiction as well as prompts for the particular work. The translator would then do a quick check to see how well those prompts have been followed, refine the prompts if necessary, and run the translation through the LLM again. These agentic models can also do basic line editing, like checking for omissions and inconsistencies. Based on the tests that I have run, I think they would do okay at first but that some problems would emerge. The results would gradually get better as the prompts and workflow are refined.
My main purpose in posting that report was not to suggest an actual business plan, even though it might be viable. Rather it was to show how powerful these most recent models are getting with higher level thinking. The space of intellectual skills that remain uniquely human continues to shrink. That's why, in my last post yesterday, I reiterated the importance of foregrounding one’s personal identity and engaging in face-to-face interaction. Show your potential customers that you are human and make them want to interact with you instead of with a machine.
Tom Gally
cpta...@ozemail.com.au
May 13, 2025, 12:02:13 AMMay 13
to hon...@googlegroups.com, SWE...@groups.io
Hi all,
Fred Uleman your ears must have been burning this weekend past. Lot’s of “when I first met Fred…” conversations.
I have just got off a Zoom call with one of my staff (literally minutes ago) where she reported on her experience interpreting for Asako Yuzuki author of “Butter” at the Melbourne Writers Festival this weekend.
Apparently one of the things they discussed was that she never thought the book would ever be translated, but the publisher put the first chapter through Google translate, put it up for auction, and Harper Collins snapped it up for a very large sum. Polly Barton then translated the whole book.
The book has now become way more popular in English than in Japanese. In Japan journalist are rather hostile and publishers kind of keeping anything feministy at arm’s length. But she is a hit in Melbourne.
Two of the highlights of IJET were Jimmy Rion railing against the existential threat of AI, and this talk by Koshimae san
https://ijet.jat.org/sessions/translating_for_publication
Where he mentions using AI for all sorts of text analysis and synonym generation with once commenting on how it might or might not threaten his business model, and also without explicitly saying so, leaving everyone pretty well reassured that there are many things which only humans can do.
Chris
--
You received this message because you are subscribed to the Google Groups "Honyaku E<>J translation list" group.
To unsubscribe from this group and stop receiving emails from it, send an email to honyaku+u...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/honyaku/CACnS3Ne-w1XOgULBNHxTiyE7oe_-CEtQ819w-o_V%3DUoQpoG8TQ%40mail.gmail.com.
Reply all
Reply to author
Forward
0 new messages