I'm a massive ChatGPT sceptic.

ChatGPT doesn't understand what it write. It just knows what everyone else would write on a topic.

So ChatGPT will say that a race should start at 11am, there will be 3 races and 1 race counts as a series of that's what it learned from the various NoR it read in the past is "normal".  Of course it also isn't going to be great at differentiating a NoR from RRS 2008-12 vs 2025-2029. Especially as it hasn't even learned on 2025 content.

If the question is - can you supply chat GPT the NoR, SI and RRS and ask if there are any discrepancy - perhaps - you can of course try it by introducing some discrepancy.

If you are saying to Chat GPT "Write me a NoR for a Laser regatta on weekend of 1st and 2nd April. Entry costs £50"... I suspect you would want to edit it a LOT... Which considering there are already NoR templates begs the question why...

But you could try it for a fictitious event and upload the prompts used and the outputs and see what people think

Some countries, like France, use standard documents, prepared by the MNA with one page for local details.
Another idea is to prepare a template with different options. The WS templates are not always up to date, and are intended mostly for dinghy events.
Here in Ireland we are preparing NoR and SI templates, including alternative wordings for common situtaions
My issues with Chat GPT would be:
- are they using up to date rules, and applying modern best practice
- how will race documents be adapted to local circumstances.

I use ChatGPT very often for daily problems, but not for RRS, NoR, SI, etc. I think there is not enough material on the internet for the chatbot to create a realistic and usable response. Often, the response is not wrong, but it is written with 'nautical words' (this is especially true for non-English questions), and you need to check the document and correct all the form errors.

----------------------------------------------------------------------------------------------------
I've tried to ask ChatGPT why the responses to my questions are sometimes not usable. This is the answer:

I understand your frustration! The main issue is that racing rules, notices of race, and sailing instructions are technical documents that are frequently updated and often use very specific terminology. I can provide answers based on my knowledge, but for precise and up-to-date information, it's always best to refer to official documents from the Italian Sailing Federation (FIV), World Sailing, or the race organizing committee.

If you point out any errors in my responses, I can improve the way I assist you. For example, I can better adapt to the correct technical language and guide you to more reliable sources when necessary. If you’d like, I can also help you interpret a notice of race or sailing instructions if you provide them!

I agree that ChatGPT probably should not be used to create NOR/SIs at this point in its evolution, although I think it could be used to point out discrepancies/compare documents. But, I definitely think it has other possible purposes in the officiating side of our sport. It is really good at making schedules and things if given specific parameters (hearing schedules?), and the newer models are good at large data base analysis (uploading many years of local wind history and the forecast of the day to find the best time to start a race?). That being said, I think that everything written by it should of course be checked, especially if it references the rules or has requirements it must meet in the RRS etc.

I have been following and testing the various AI tools floating around since their introduction.

None of them are appropriate for highly structured documents like SIs and NoRs. Other types of users have come to the same conclusion. The reason is that ChatGPT (and all the others) were never intended to create highly structured and carefully unambiguous language. Their goal is to write in a way that most people recognize and might think was performed by a human.

Neither SIs nor NoRs are targeted at writing in this way. My view is that a NoR is a contract between a competitor and an Organizing Authority, as well as a solicitation for competitors to participate. I do not believe that the NoR should attempt to be a piece of marketing literature. That should be a separate document provided to competitors. As such, the exact terms of art are critical and deviating from the sample in RRS Appendix J is risky. Most of the problems that I've had to solve were created by a human trying to be "creative" with the language rather than adhering to Appendix J.

This is not to say that Appendix J covers every issue. Additions based on specific regulatory, environmental, and physical factors need to be included. But altering Appendix J's language is most often a serious mistake. 

I agree with all the comments above. AI doesn't have any concept of "consistent" unless trained in specific data sets to develop consistent documents in that particular context.  It's also not clear what you would gain by using a generalized model like ChatGP -- you'd have to provide it with all the specific information, like number of races, dates and times, etc., and by the time you did that you could have simply filled out the templates in Appendices KG and NG.  Alternatively, you could use Appendix S for your sailing instructions, though I haven't seen that done for many years.

World sailing has standard documents for this and some MNAs have translated ones as well. I don’t see the need to use AI for those documents. 

> "tasks like reviewing these documents for consistency and ensuring compliance with RRS...  are areas where AI tools like ChatGPT could excel."
If you want really critical review of your documents, just offer them up right here at "chatRRS."  :-)

You made me curious what it might produce. I don't pay for ChatGPT so I think I have an older model.

To the prompt:
Write me a notice of race for a sailing event

It gave something not too dissimilar to the World Sailing templates. I didn't go through it line by line. But similar idea, and nothing immediately looked wrong. But lots of [enter XX here].

It asked "would I like specific customisation for my event" and so I played along and said "yes".

It gave me a list of things to consider. I told it some dates but nothing more. Just doing that, it started to hallucinate information. I was still getting a template for changes. But it decided that the 3 day even I suggested, day 1 was for registration & briefing. 

It then asked if I wanted to fill in the missing details or for ChatGPT to do it. So being cheeky I said he should do it, but didn't give any details!

I still got some [fields] to complete. But it started to get stuff wrong. 


2. RULES

2.1 The regatta will be governed by:

3.1 The event is open to the following classes:

I'm in the UK, so why it would set prices in dollars I don't know. Not very intelligent ;-). But now we have AI picking my event costs, picking boat classes,  using last year's rules.


8. COURSES

Courses will be a combination of windward-leeward and triangle configurations for dinghies and coastal courses for keelboats. Final details will be included in the sailing instructions.


And it's picked the courses 

And let's hope the weather is good - because I have to get 3 races in to even get people a result. (The discard language is not standard)

Each participating boat must have valid third-party liability insurance with a minimum coverage of $500,000 (or equivalent in local currency).


That insurance is very low.

Anyway. I decided to ask it to write the accompanying SI...

We stayed with the 2021-24 RRS.
The courses no longer included triangles. 
Oddly there was a whole section on recalls.
It picked race durations etc for me. That's the kind of thing if you didn't notice could be problematic (90mins).

But now the SI conflicted with the NoR and said. 1 race to count as a series.


14. PENALTIES

14.1 RRS 44.1 applies, with a Two-Turns Penalty for rule breaches.
14.2 For keelboats, the penalty may be a One-Turn Penalty, at the discretion of the jury.



16. SAFETY REGULATIONS

16.1 All competitors must wear personal flotation devices (PFDs) at all times while afloat, except briefly while changing clothes.

Pretty sure that needs a 'this amends rule 40'

So I then asked it to check for discrepancies.


4. SCHEDULE OF RACES

❌ Discrepancy:

Since it's a 90 minute race, the maths doesn't add up!!




5. SCORING

❌ Discrepancy:

Why would you need things twice?.this verbage however is closer to normal.

It missed the 3 vs 1 for a series

✅ No major discrepancies. However, if different courses apply to different fleets (e.g., windward-leeward for dinghies, coastal for keelboats), the NoR should explicitly state this for clarity.

It missed that the triangles are gone!!


7. TIME LIMITS

✅ No discrepancy. The SIs specify a 90-minute time limit per race, which aligns with standard practice.


Standard practice? According to who? It decided this was an Oppi race... We are going to break some kids!

I asked it if the last warning signal was at 3pm when the prize giving could be. It told me the last race would finish at 4 because the race time was 90 minutes... But it then understood to protest time limit and said that means 5pm.at the earliest. 

To be fair it does then say "would you like me to update the documents"  ... But it's also wrong!!

Calum, thank you. You (and ChatGPT) made my day. As a judge, I frequently deal with notices of race and sailing instructions that are not well written, but the errors that ChatGPT makes are much more ... inspired?

In a related way, there is a great use of AI that doesn't "write" the documents for you, but it can help you figure out where to look in all the various documents that already exist for examples and most importantly for Cases that interact with specific term and situations to inform the correct use of all the various sources.

It is called "NotebookLM" and can be found here: https://notebooklm.google/

A good friend, who is a tremendously successful navigator, vectored me to it. I have entered all of the current versions of every World Sailing and US Sailing document I can find. I called it "Racing Rules of Sailing" which is ambiguous (duh) but helpful to me. NotebookLM has scanned all the files and can answer questions complete with links to all the places it found the answers. That last bit is critical, as NotebookLM makes mistakes, so before I use an "answer", I check out the correct source documents.

This has been tremendously helpful when drafting NoRs and SIs.

Here is what NotebookLM had to say about what I built:

"The provided texts encompass various aspects of sailing regulations and their interpretation, offering guidance for both competitors and race officials. Several documents, including the Racing Rules of Sailing and associated Case Books, clarify the application of specific rules through examples and scenarios, addressing topics like right-of-way, penalties, mark roundings, and protests. There are also Call Books tailored for team and match racing formats, providing specific umpire guidance. Additionally, some sources detail rules and procedures specific to windsurfing slalom, while others outline changes and updates to the racing rules effective in 2025-2028 and provide guidance for race management and judging."

I entered this question to NotebookLM: "How do I change the size of the zone around a mark?"

NotebookLM had this to say in response to my question:

"Based on the sources, the size of the zone around a mark is defined within specific sets of racing rules. Here's what the sources indicate:

It's important to note that the provided sources do not give a definitive answer on whether the sailing instructions for a regular event can alter the size of the zone. However, they do show that the zone size can differ between specific rule variations."

BTW: I have yet to find a text version of the Racing Rules, but have stuffed the PDF into NotebookLM to see if it'll hack through it and understand it.

One more NotebookLM example:

Question: "Which cases apply to tacking in the zone?"

Answer:

"Based on the sources, the following cases and rules apply to tacking in the zone around a mark:

I wonder how ChatGPT would do reconciling last year's NOR and SI with this year's template. I will try to find out at some point this spring.