Hello,
I was fiddling with AI updating my resume and was so impressed by the content and improvements it provided that I decided to give it a go with modelling subjects.
So were do I start? Here are the preliminary steps I followed:
I selected the 1934 J Class Endeavour as the subject of research so the first thing to add to the task description was specifying the subject:
The next essential step is to let the AI know who it shall impersonate:
The aim was on collecting information about available plans kits, books, building logs and the like with the scope of having plans for scratch building and some guidelines so we specify our expectations. I think this step needs refinement as it contains too many items:
In order to guide the process in the direction we want, we shall specify a few steps the AI shall follow thus narrowing the options and improving the out quality. Again, these steps are essential and shall probably be further improved:
Once this is done, we shall get more specific on what we expect again improving the generated content:
Positive and negative example are one more step to quality content:
The finally, it's time to describe the expected deliverables:
Finally, it was time to run the above by Claude AI (https://claude.ai/). The result was ready in about 2-3 minutes and could be exported as a PDF file.
I'm attaching both a text file containing the steps above (ShipModelling_AI.txt) and the first result as a PDF (Claude AI Generated - 1934 J-Class Endeavour Scale Model Building Guide.pdf).
Microsoft Copilot was not dealing well with the input and I was not able to make it generate more than a few sentences so I gave up on that for now. Claude AI, on the other hand was performing quite well (see attached PDF for the result).
I realise, this is very rough for now as I only started and am no expert by any means so your input is more than welcome here! Here are a few ideas I'd like to toss in here in case you care to contribute:
I was fiddling with AI updating my resume and was so impressed by the content and improvements it provided that I decided to give it a go with modelling subjects.
So were do I start? Here are the preliminary steps I followed:
- Select an AI to use: I went with Microsoft Copilot and Claude AI, manly because I used these previously
- I think it is well worth checking other AI options if you know any that performs well
- Find a template that can be used for the so called prompt, the task description you gave the AI: a colleague suggested an open source option called Fabric (https://github.com/danielmiessler/fabric/)
- I'm not using the tool itself but the patterns they provide look like a solid starting point for formulating the task
I selected the 1934 J Class Endeavour as the subject of research so the first thing to add to the task description was specifying the subject:
Code:
The vessel under analysis is the 1934 J Class Endeavour.
The next essential step is to let the AI know who it shall impersonate:
Code:
You are an advance AI expert on scale modelling that specialises in creating building plans for wooden ship models in scales in the range of 1:20 to 1:200.
The aim was on collecting information about available plans kits, books, building logs and the like with the scope of having plans for scratch building and some guidelines so we specify our expectations. I think this step needs refinement as it contains too many items:
Code:
The goals of this exercise are to:
Create a list of available kits or starter packs for the specified vessel including links to the kit on the manufacturer's website.
Create a list of available building logs from forums, videos, magazines and books for the specified vessel. Include links to each of the entries. Provide a short summary of the content no longer than 200 words and content rating on a scale from 0 to 5.
Create ready-to-use scale modelling plans based on the most details line drawings found in your research. These plans shall be easily scaleable up or down and printable on home printers of commercial printing services.
Create a list of materials required for building the model based on the provided plans.
Create an article that walks an inexperienced builder through the model building process based on the plan created above.
Create a list vessel that have similarities to specified one that may be of interest to the user.
In order to guide the process in the direction we want, we shall specify a few steps the AI shall follow thus narrowing the options and improving the out quality. Again, these steps are essential and shall probably be further improved:
Code:
STEPS
Start by slowly and deeply consuming the input you've been given. Re-read it 218 times slowly, putting yourself in different mental frames while doing so in order to fully understand it.
Create a 10 meter by 10 meter whiteboard in your mind, and write down all the different information from what you read. You will continuously update this whiteboard as you discover new insights.
Thoroughly research the specified vessel using publicly available sources including libraries, books, magazines, blogs, videos and forums on ship real building process from the 16th century to current times. Aim to understand the main concepts that can be translated to a static scaled down version of a ship model.
Research publicly available sources on ship model building focusing on building a model from scratch based on available historical and ship manufacturer plans. The main building material to be considered is commercially commonly available wood in Europe. Small details can also be manufactured from metals like copper. Avoid the use of tools that are either expensive or take up a lot of space.
Research wooden model ship kit manufacturer offerings and extract useful product design ideas and part split strategies.
Research wooden model ship building forums and extract a list of useful building and painting techniques.
Now step back and look at the entire whiteboard, and the entire situation in your mind again. Look at all the stuff you have on the board so far, and reconsider everything you've learned again, and then enhance the whiteboard diagram with any new insights you find. Update the whiteboard.
Perform these steps 100 times, optimising on each iteration.
Once this is done, we shall get more specific on what we expect again improving the generated content:
Code:
OUTPUT
Create a list of available resources free and commercial about the specified vessel including links.
Create a list of available building plans, separating the list in free and commercial offerings for the specified vessel. Include links to the source.
Create a list of available kits or starter packs for the specified vessel including links to the kit on the manufacturer's website.
Create a list of available written building logs from forums, magazines and books for the specified vessel including links.
Create a list of available video building logs for the specified vessel, providing a short summary of the video content no longer than 200 words and content rating on a scale from 0 to 5.
Provide ready-to-use scale modelling plans based on the most details line drawings found in your research. Include links to available plans. These plans shall be easily scaleable up or down and printable on home printers of commercial printing services.
Create a list of materials required for building the model based on the provided plans.
Create an article that walks an inexperienced builder through the model building process based on the plan created above.
Create a list vessel that have similarities to specified one that may be of interest to the user
Positive and negative example are one more step to quality content:
Code:
POSITIVE EXAMPLES
Building plans that use the on plank on bulkhead (POB) method
Building plans that use commonly available hand tools.
NEGATIVE EXAMPLES
Building plans that require a fully equipped workshop with expensive machines like a lathe or plank planner
The finally, it's time to describe the expected deliverables:
Code:
OUTPUT INSTRUCTIONS
Output text in Markdown, but don't use bold or italics because the asterisks are difficult to read in plaintext.
Output drawings as vector graphics. Where vector graphics are not feasible, use high resolution, high quality raster images.
Finally, it was time to run the above by Claude AI (https://claude.ai/). The result was ready in about 2-3 minutes and could be exported as a PDF file.
I'm attaching both a text file containing the steps above (ShipModelling_AI.txt) and the first result as a PDF (Claude AI Generated - 1934 J-Class Endeavour Scale Model Building Guide.pdf).
Microsoft Copilot was not dealing well with the input and I was not able to make it generate more than a few sentences so I gave up on that for now. Claude AI, on the other hand was performing quite well (see attached PDF for the result).
I realise, this is very rough for now as I only started and am no expert by any means so your input is more than welcome here! Here are a few ideas I'd like to toss in here in case you care to contribute:
- try other (free) AI options
- different prompting methods
- the subject research steps can be improved by providing precise guiding in form of specific resources
- provide a list of kit manufacturers include more obscure ones
- include links to reference materials like plan interpretation, material selection, tutorials of any kind