Is AI a tool that can aide boaters?

[jssailem]

Last September, I posted this message wondering how long it would be before AI intruded on my sailing.

Just today, I discovered that AI is here. It has a beginning price tag of 10,000 euro

SEA.AI Watchkeeper: designed for Motorboats and Sailing Yachts

Discover SEA.AI Watchkeeper, the AI thermal camera with wide-angle detection, floating object alerts & smart features for boaters.

www.sea.ai

 

[Davidasailor26]

Less than two years after this thread was started I have a sailing related AI use case that would’ve been pretty unimaginable then - I spent an evening prompting Gemini, and “vibe coded” a web application to use as a logbook. With one summary prompt to get started and maybe 10 prompts to revise and debug what it built I have a working website that can pretty much replace my older spreadsheet logs. There are a few more features I’d like to add. One will be a function to query the log in natural language for questions like “when did I last clean my winches” or “how many miles have I gone so far this year compared to my average”. But for a couple hours of tinkering without touching any actual code myself this is a pretty interesting start.

 

[jssailem]

Thank you for triggering me to review what I posted in 2024. A lot has changed. AI is empowering individuals, with an understanding of what they want to accomplish, with the tools to write code to create new tools. Like having a smart computer coder in your back room (iPhone/Laptop) to write the APP to work on the hardware you possess.

Now that is pretty cool, and what an advancement in 19 months.

How tight was the code? Did you review the code? How did it differ from what you would have written?

 

[Davidasailor26]

How tight was the code? Did you review the code? How did it differ from what you would have written?

I haven’t looked it over yet. This was literally just a couple hours of intermittent prompting while I watched TV. I think as I try to add more functionality I’ll be able to tell whether the foundation was good enough to be modular and extensible, or if it’s all spaghetti that needs to get rewritten for each new change. From other pure AI code I’ve seen I certainly don’t expect it to be perfect. But for a quick personal utility or a prototype that could be good enough.

 

[Davidasailor26]

Here’s another interesting use I don’t think we really reconized when this thread started - I took all of the manuals I had as PDF’s for things like my engine and instruments and put them in a Google Notebook LM. With that I can use Google’s AI to ask natural language questions like “what oil should I use for my transmission”, or “what’s the displacement of my boat”. The answers include citations to the original text so I can see it’s not hallucinating. I also asked it questions like “what is the color code for the paint on my engine”, that I know isn’t in the files. It correctly told me it didn’t know, no random guessing, and it suggested that since it’s a Yanmar the common Yanmar Gray could be a good choice for touch-ups.

 

[dlochner]

Here’s another interesting use I don’t think we really reconized when this thread started - I took all of the manuals I had as PDF’s for things like my engine and instruments and put them in a Google Notebook LM. With that I can use Google’s AI to ask natural language questions like “what oil should I use for my transmission”, or “what’s the displacement of my boat”. The answers include citations to the original text so I can see it’s not hallucinating. I also asked it questions like “what is the color code for the paint on my engine”, that I know isn’t in the files. It correctly told me it didn’t know, no random guessing, and it suggested that since it’s a Yanmar the common Yanmar Gray could be a good choice for touch-ups.

So, AI becomes a personalized search engine on a closed finite data set. A handy use and time saver.

 

[Stu Jackson]

on a closed finite data set.

Very astute observation.
It appears from my wide-ranging reading on this AI "issue" that AI has extremely excellent results when the SOURCE material is well defined and "vetted" as reliable. Open sourced, unverified material that an AI engine may use in response to a question remains as questionable as the material it used, and should be suspect to "trust but verify," just like any manual search one could perform.
That's why it is so good at coding, math and medical, but questionable in, say, boating. I'm sure medical AI doesn't use Web MD as source.

 

[dlochner]

That's why it is so good at coding, math and medical, but questionable in, say, boating.

In the medical world there is often considerable symptom overlap, a cluster of symptoms can be diagnostic of more than one disease. My fake heart attack last summer was a good example of that.

 

[jssailem]

but questionable in, say, boating.

Perhaps.

Maybe it is not the subject application, but a better understanding of the tool and how to harness its power.

Here is an example. In my youth, Arvel Gentry was a Boeing Engineer and Sailor. He challenged the conventional wisdom, which was my curriculum, from successful racing sailors in Southern California.

Arvel Gentry (1933-2015) was an aerodynamicist. He had a plentiful professional career, including having been involved in aerodynamics research at the biggest aircraft companies in the world, like NACA, Lockheed, Douglas, McDonnell Douglas, and Boeing. His name rose to sailing-fame on the early 1970's when he published many sail aerodynamic articles aimed at the average sailor, and changed sailing forever. His approach to sailing from a scientific point of view started an evolving process whose outcomes we see now everywhere, both in the aero and hydrodynamic fields of sailing.​

​

Using Gentry as a reference point, I asked Claude and Grok (2 popular AI programs) to answer this sailing question:

Using Arvel Gentry's research on sails and boat racing summaize the trimming of a jib and main sail for optimum performance in a 10 knot and 15 knot wind when there is a 3-foot sea running towards the direction of the boat.​

The response took less than 60 seconds to prepare in each system. I believe it was instructive.

Gentry's Jib-Main Trim in 10-15 Knots with Chop | Shared Grok Conversation

Using Arvel Gentry's research on sails and boat racing summaize the trimming a jib and main sail for

grok.com

Claude

Shared via Claude, an AI assistant from Anthropic

claude.ai

 

[Stu Jackson]

Maybe it is not the subject application, but a better understanding of the tool and how to harness its power.

Using Gentry as a reference point,

John, doesn't your input here just reinforce my point?

has extremely excellent results when the SOURCE material is well defined and "vetted" as reliable.

Thanks. You used a very good example, too.
Interestingly, I have found that Arvel's work is excellent and "usable" although "heavy" in verbiage. The most useful Sail Trim material I have found after reading all of Sail magazine's material and their two books (often at odds from article to article within!), is Don Guilette's Sail Trim Users Guide available right here on this very website: https://shop.sailboatowners.com/prod.php?51998

 

[mermike]

I vibe coded a boat selection app based on Analytical Hierarchy Process. Took about an hour to create a functional, standalone web app.

Then i had it code an app to capture and edit settings from an Arco Zeus voltage regulator, so I could edit from the desktop.

Then i uploaded ABYC standards E-10 and E-11. I added a PDF of an electrical schematic I drew. It spit out a circuit schedule that was about 80% complete and told me where it had questions about the fusing.

This week I told Claude I was driving to California from Boise and wanted a route with some primitive camping along the way. We stopped at Walker Lake and just pulled in to Alabama Hills. Both good suggestions.

The more I use it the more freaked out I am.

 

[jssailem]

Stu. It is about perspective. I am not ready to let AI run my life, nor tell me how to sail my boat. I feel like I did in early 1985 about computer solutions to mathematical problems. Everyone was jumping on board with Visicalc, Lotus 123, etc. Most had forgotten how to calculate mathematical formulae. They would build a spreadsheet and put in some data and numbers, then declare that because the computer produced an answer, it was the gospel truth.

When Google did a search for data, the solutions must be factual in the late 90's.

Nowadays, we know those were fallacies. AI is in it's inface. It is far more powerful than either of those apps. It has way more capacity to sort through data and assemble it. The quickness of the process leads to a bias that the answer must be valid. This is a false result.

On the other hand, we can recognize such limits (biases) and harness them to find useful solutions.