3 ms·

Is this idea of automatically generating documents has potential?

2 points by xqliu 3y ago

xqliu 3y ago
The whole situation is like this, Our team had some accumulations in requirement management, and AI, and then, following the trend of AI, we wanted to create something. We developed a service at supadoc.ai This service uses AI to automatically generate documents for users. Currently implemented features include: Marketing plan Software system design Product design Business plan Product test cases Would you please please help to take a look and let us know if you have any need for this? It's been nearly half a month since launch, and after my tireless promotion on Twitter, we've got around 700 visitors to the website and around 50 of users, but no paying ones. Initially, our plan was to cater to roles like product managers, project managers, and testers. However, after some time in practice, we've found that these types of users don't necessarily have valid use cases. Therefore, we are now shifting our focus towards people who need to prepare formatted Word documents. This type of Word document doesn't require high precision, but it often takes several steps to generate the document they need, which can be quite time-consuming. Our tool's main initial goal is to "improve efficiency", and we are also exploring how to provide "added value" to our users in the later stages. So now we are wondering whether this really has a future. Would you use this tool? If you could provide your thoughts, we would be very grateful. It would be a ray of light in our darkness~
cebert 3y ago
What would be the value proposition of using the service over using GPT-4 directly? I don’t see a lot of value add here.
- xqliu 3y ago
  Yeah, that's a really good question, I think the tiny value we could offer is 1. Help to do a chain of request(around 10 to 20) and transfer a short description to a long document. this could save a lot of hand work for the user, for example, the https://www.supadoc.ai/samples/Marketing%20Plan%20EN.docx https://www.supadoc.ai/samples/Marketing%20Plan%20EN.docx was generated from the follow description "This is a SAAS product for the System Engineer. The System Engineer will input some description. The product will use the latest AI technology to analysis the description. Then the product will generate one Requirement Document according to AI analysis data. The SAAS platform needs to be promoted now. A Marketing plan is required." User can also benefit from the prebuild best practice document structure we build into the system. 2. We could help to generate good shape docx document with TOC, headers etc and also saves user some time. But overall, We also think that the logic behind our product is not so solid, that's not a killer app and can only make user more productivity, instead of kill user's pain. Thanks for the reply, we are also discussion the direction internally, do you have any insights which you may share with us? Sincerely appreciate again :)
  - cebert 3y ago
    I don’t know how feasible it is, but I imagine larger orgs like the one I work in have document templates “e.g Business Architecture Review and Incident Postmortem”. I could see it being helpful to be able to inform an AI tool of these templates and have it assist you will filling out the sections of the template.
    - xqliu 3y ago
      Yeah, We also think that's a valid scenario, but seems as of now we are marketing on a 2C community and world... That's something we definitely need to figure out how to fix the gap between marketing and target customer... Initialise it's also a requirement raised by one of our big company customer. BTW: I would like to share more about the reason behind Why we are trying to build http://supadoc.ai http://supadoc.ai? Our team is confident that GPT-4's capabilities have reached a point where it can serve as the primary author of documents. This goes beyond the current trend in collaborative software, where AI is used only for minor, isolated adjustments without a comprehensive view. This understanding forms the basis of our approach. With this in mind, we aim to identify fitting scenarios through a process of rapid iteration and embracing quick failures. Moreover, we intend to leverage this method of applying technology as a way to promote our low-code platform on http://muyan.io http://muyan.io, which is our main area of focus currently. Thanks for your reply and info, and wish you a good day
oldandtired 3y ago
About 15 or so years ago, I was involved in a project to do automatic document generation. The documents that were to be generated were for various clients in various industries. I won't name the company for various reasons. The idea at the time was to produce a customised document based on client, client industry, products applicable and a few other specified (but varying requirements). The customisation of the document was in the hands of the relevant sales/marketing person involved and was based a that person specifying a range of options (standardised by industry/products). What became obvious over the time of developing this particular systems was that the sales/marketing people were actually not interested in such a facility, even though it would save them many hours producing the documents required. Work undertaken with other companies that had interactions with sales/marketing teams also highlighted that there is a level of distrust with such systems. Over many years, I have had to write user guides, functional specifications, technical specifications and product documentation and even though there may be company standards for such documentation, every document has had much that was not common even in the same class of documents. I don't know if your system would be that helpful to many people. It might be useful for a small class of people. However, I don't think that this would be a useful idea to try and sell, so to speak. Of course, things may have changed in the last ten years.
- xqliu 3y ago
  Dear, Really appreciate for your sharing. Your experience is exactly what we were thought about .... Over many years, I have had to write user guides, functional specifications, technical specifications and product documentation and even though there may be company standards for such documentation, every document has had much that was not common even in the same class of documents. ---> I think here is where the new LLM can help, it actually can help on prepare most of the contents and build best practice into the document. I don't know if your system would be that helpful to many people. It might be useful for a small class of people. However, I don't think that this would be a useful idea to try and sell, so to speak. ---> Yeah you are absolutely right, we also think that the customer base for such a system is not large. So truth.. though not sounds great.. Really appreciate. We are also thinking about a new approach to help user write doc using AI, by let user to # Create document 1. Input a description 2. GPT to generate a outline based on the description 3. User can interactive with the generated outline, like restructure the outline, change outline title, add / remove outline items etc. 4. And then we will let GPT generating the actual content of each header. # Edit document 1. User can interactive with the generated outline, like restructure the outline, change outline title, add / remove outline items etc. 2. User can input description for instruction, we will combine it together with the structure changes user made on step 1. 3. Then we will ask GPT to generate the actual content of each header. We think that if we can follow the above steps, together with the old content and structure of the whole doc, all send to GPT, GPT will do a great job to write / edit / rewrite the doc, with contents which actually is correct and usable. But also concerning about the token consume... it looks like it will uses many tokens if we really want GPT to have a complete high level insight of the doc and can do a proper adjustment for sub level header. Do you have any suggestion on this scenario? Thanks again for your great sharing :)
xqliu 3y ago
Another approach we are thinking about is to 1. Let user to upload a semi finished / finished doc 2. Let user input a instruction about the changes 3. Our system send the instruction and the original doc to GPT together, and let GPT to change the doc based on user's instruction. 4. Return back the updated doc file to user. Do you think this approach will work and bring more value to users? Thanks all for your input :)

2 points by xqliu 3y ago

The whole situation is like this, Our team had some accumulations in requirement management, and AI, and then, following the trend of AI, we wanted to create something. We developed a service at supadoc.ai This service uses AI to automatically generate documents for users. Currently implemented features include: Marketing plan Software system design Product design Business plan Product test cases Would you please please help to take a look and let us know if you have any need for this? It's been nearly half a month since launch, and after my tireless promotion on Twitter, we've got around 700 visitors to the website and around 50 of users, but no paying ones. Initially, our plan was to cater to roles like product managers, project managers, and testers. However, after some time in practice, we've found that these types of users don't necessarily have valid use cases. Therefore, we are now shifting our focus towards people who need to prepare formatted Word documents. This type of Word document doesn't require high precision, but it often takes several steps to generate the document they need, which can be quite time-consuming. Our tool's main initial goal is to "improve efficiency", and we are also exploring how to provide "added value" to our users in the later stages. So now we are wondering whether this really has a future. Would you use this tool? If you could provide your thoughts, we would be very grateful. It would be a ray of light in our darkness~

cebert 3y ago

What would be the value proposition of using the service over using GPT-4 directly? I don’t see a lot of value add here.

xqliu 3y ago
Yeah, that's a really good question, I think the tiny value we could offer is 1. Help to do a chain of request(around 10 to 20) and transfer a short description to a long document. this could save a lot of hand work for the user, for example, the https://www.supadoc.ai/samples/Marketing%20Plan%20EN.docx https://www.supadoc.ai/samples/Marketing%20Plan%20EN.docx was generated from the follow description "This is a SAAS product for the System Engineer. The System Engineer will input some description. The product will use the latest AI technology to analysis the description. Then the product will generate one Requirement Document according to AI analysis data. The SAAS platform needs to be promoted now. A Marketing plan is required." User can also benefit from the prebuild best practice document structure we build into the system. 2. We could help to generate good shape docx document with TOC, headers etc and also saves user some time. But overall, We also think that the logic behind our product is not so solid, that's not a killer app and can only make user more productivity, instead of kill user's pain. Thanks for the reply, we are also discussion the direction internally, do you have any insights which you may share with us? Sincerely appreciate again :)
- cebert 3y ago
  I don’t know how feasible it is, but I imagine larger orgs like the one I work in have document templates “e.g Business Architecture Review and Incident Postmortem”. I could see it being helpful to be able to inform an AI tool of these templates and have it assist you will filling out the sections of the template.
  - xqliu 3y ago
    Yeah, We also think that's a valid scenario, but seems as of now we are marketing on a 2C community and world... That's something we definitely need to figure out how to fix the gap between marketing and target customer... Initialise it's also a requirement raised by one of our big company customer. BTW: I would like to share more about the reason behind Why we are trying to build http://supadoc.ai http://supadoc.ai? Our team is confident that GPT-4's capabilities have reached a point where it can serve as the primary author of documents. This goes beyond the current trend in collaborative software, where AI is used only for minor, isolated adjustments without a comprehensive view. This understanding forms the basis of our approach. With this in mind, we aim to identify fitting scenarios through a process of rapid iteration and embracing quick failures. Moreover, we intend to leverage this method of applying technology as a way to promote our low-code platform on http://muyan.io http://muyan.io, which is our main area of focus currently. Thanks for your reply and info, and wish you a good day

oldandtired 3y ago

About 15 or so years ago, I was involved in a project to do automatic document generation. The documents that were to be generated were for various clients in various industries. I won't name the company for various reasons. The idea at the time was to produce a customised document based on client, client industry, products applicable and a few other specified (but varying requirements). The customisation of the document was in the hands of the relevant sales/marketing person involved and was based a that person specifying a range of options (standardised by industry/products). What became obvious over the time of developing this particular systems was that the sales/marketing people were actually not interested in such a facility, even though it would save them many hours producing the documents required. Work undertaken with other companies that had interactions with sales/marketing teams also highlighted that there is a level of distrust with such systems. Over many years, I have had to write user guides, functional specifications, technical specifications and product documentation and even though there may be company standards for such documentation, every document has had much that was not common even in the same class of documents. I don't know if your system would be that helpful to many people. It might be useful for a small class of people. However, I don't think that this would be a useful idea to try and sell, so to speak. Of course, things may have changed in the last ten years.

xqliu 3y ago
Dear, Really appreciate for your sharing. Your experience is exactly what we were thought about .... Over many years, I have had to write user guides, functional specifications, technical specifications and product documentation and even though there may be company standards for such documentation, every document has had much that was not common even in the same class of documents. ---> I think here is where the new LLM can help, it actually can help on prepare most of the contents and build best practice into the document. I don't know if your system would be that helpful to many people. It might be useful for a small class of people. However, I don't think that this would be a useful idea to try and sell, so to speak. ---> Yeah you are absolutely right, we also think that the customer base for such a system is not large. So truth.. though not sounds great.. Really appreciate. We are also thinking about a new approach to help user write doc using AI, by let user to # Create document 1. Input a description 2. GPT to generate a outline based on the description 3. User can interactive with the generated outline, like restructure the outline, change outline title, add / remove outline items etc. 4. And then we will let GPT generating the actual content of each header. # Edit document 1. User can interactive with the generated outline, like restructure the outline, change outline title, add / remove outline items etc. 2. User can input description for instruction, we will combine it together with the structure changes user made on step 1. 3. Then we will ask GPT to generate the actual content of each header. We think that if we can follow the above steps, together with the old content and structure of the whole doc, all send to GPT, GPT will do a great job to write / edit / rewrite the doc, with contents which actually is correct and usable. But also concerning about the token consume... it looks like it will uses many tokens if we really want GPT to have a complete high level insight of the doc and can do a proper adjustment for sub level header. Do you have any suggestion on this scenario? Thanks again for your great sharing :)

xqliu 3y ago

Another approach we are thinking about is to 1. Let user to upload a semi finished / finished doc 2. Let user input a instruction about the changes 3. Our system send the instruction and the original doc to GPT together, and let GPT to change the doc based on user's instruction. 4. Return back the updated doc file to user. Do you think this approach will work and bring more value to users? Thanks all for your input :)