I went out a couple of weeks ago on the tail of a typhoon.
Not to far and a fun ride.
However on the way back in the Coast Guard (small ) cutter met us and hailed us.
Told us it was a no no to be out in those conditions.
The port was full of (large) fishing vessels.
we had planned to go out yesterday, but there was an earthquake in Kamchatka.
So I bailed. I didn't want another run in with the Coast Guard!
One of my crew consulted Chat GTP about the safety of going out.
This was the reply.
Interesting strategy I tried today: I asked ChatGPT for a summary and got a beautifully concise account of the whole thing.
Well, apparently, Kamchatka just had the 6th biggest quake in the world ever.
Aftershocks expected for the next 5 weeks or so.
Mixed news from the coast: TV says big damage to aquaculture (shellfish?), & no reason not to believe that. Apparently damage in the Sendai Bay area, too. The waves might not be much to look at but that's deceptive for what goes on under water.
Asked about risks, ChatGPT advises that:
tsunami advisories have been lifted by Japan Met. Agency
sailing should be possible
there may be residual surges & strong currents but should be mostly subdued by tomorrow
aftershocks should not cause problems around here
proceed with caution if you go out, & keep in touch with latest news
. . . if you trust ChatGPT (I'm using a freebie version) . . .
Cheers
Ian
We just finished a ton of development on ChatGPT. We were trying to improve its accuracy and attention span. Some nuances worth knowing about....
a) It's a loop. After 2-3 hours (and I'm using the paid version) it has no memory of your chat. You may be able to see the history when you scroll up, but ChatGPT has long forgotten what it wrote or decided. Don't assume it knows something just because it gave you an answer a few hours ago.
b) You have to give it rules and (important) a role: "You (Chatgpt) are a navigator on a sailing vessel. Use only the polar diagrams I have uploaded to you to make routing decisions. Use Australian English for grammar, dates and units of measure. "
c) Dont be nice, talk like a grumpy old man. Words like please and thank you increases the number of tokens and slows down the chat thread the longer it goes on.
d) Have lots of smaller chat sessions, rather than one big chat session. I have a ChatGPT project called 'race car'. In that project I have seaparate chats for each discipline, eg: engine, fuel injection, ignition system, rear suspension, chassis tuning, etc.
To get it really clinically accurate, we ended up creating a virtual clone of the real world. We wrote narratives describing virtual :
- federal, state and local governments agencies;
- industry standards bodies and standards;
- (own) company;
- human roles (governement tender evaluator, QA auditor, OH&S officer).
- lists of allowable/restricted information sources, rules,
- and on and on it goes.
I now have a ChatGPT template that provided amazingly accurate responses. But I have 20 x odd text files with a nuts amount of detail establishing our chat 'rules' to get it, and keep it, on point.
I have found similar with software development. I now create a "context" file in order to persist the aims and parameters between sessions.
Select to expand quoteshaggybaxter said..
We just finished a ton of development on ChatGPT. We were trying to improve its accuracy and attention span. Some nuances worth knowing about....
a) It's a loop. After 2-3 hours (and I'm using the paid version) it has no memory of your chat. You may be able to see the history when you scroll up, but ChatGPT has long forgotten what it wrote or decided. Don't assume it knows something just because it gave you an answer a few hours ago.
b) You have to give it rules and (important) a role: "You (Chatgpt) are a navigator on a sailing vessel. Use only the polar diagrams I have uploaded to you to make routing decisions. Use Australian English for grammar, dates and units of measure. "
c) Dont be nice, talk like a grumpy old man. Words like please and thank you increases the number of tokens and slows down the chat thread the longer it goes on.
d) Have lots of smaller chat sessions, rather than one big chat session. I have a ChatGPT project called 'race car'. In that project I have seaparate chats for each discipline, eg: engine, fuel injection, ignition system, rear suspension, chassis tuning, etc.
To get it really clinically accurate, we ended up creating a virtual clone of the real world. We wrote narratives describing virtual :
- federal, state and local governments agencies;
- industry standards bodies and standards;
- (own) company;
- human roles (governement tender evaluator, QA auditor, OH&S officer).
- lists of allowable/restricted information sources, rules,
- and on and on it goes.
I now have a ChatGPT template that provided amazingly accurate responses. But I have 20 x odd text files with a nuts amount of detail establishing our chat 'rules' to get it, and keep it, on point.
Sounds like trying to teach your son how to mow the lawn.
Select to expand quotecammd said..shaggybaxter said..
We just finished a ton of development on ChatGPT. We were trying to improve its accuracy and attention span. Some nuances worth knowing about....
a) It's a loop. After 2-3 hours (and I'm using the paid version) it has no memory of your chat. You may be able to see the history when you scroll up, but ChatGPT has long forgotten what it wrote or decided. Don't assume it knows something just because it gave you an answer a few hours ago.
b) You have to give it rules and (important) a role: "You (Chatgpt) are a navigator on a sailing vessel. Use only the polar diagrams I have uploaded to you to make routing decisions. Use Australian English for grammar, dates and units of measure. "
c) Dont be nice, talk like a grumpy old man. Words like please and thank you increases the number of tokens and slows down the chat thread the longer it goes on.
d) Have lots of smaller chat sessions, rather than one big chat session. I have a ChatGPT project called 'race car'. In that project I have seaparate chats for each discipline, eg: engine, fuel injection, ignition system, rear suspension, chassis tuning, etc.
To get it really clinically accurate, we ended up creating a virtual clone of the real world. We wrote narratives describing virtual :
- federal, state and local governments agencies;
- industry standards bodies and standards;
- (own) company;
- human roles (governement tender evaluator, QA auditor, OH&S officer).
- lists of allowable/restricted information sources, rules,
- and on and on it goes.
I now have a ChatGPT template that provided amazingly accurate responses. But I have 20 x odd text files with a nuts amount of detail establishing our chat 'rules' to get it, and keep it, on point.
Sounds like trying to teach your son how to mow the lawn.
Nyuk nyuk nyuk!
Select to expand quotecammd said..
Sounds like trying to teach your son how to mow the lawn.
Is it too soon to call that one the most accurate post of the thread?
Timely thread I was thinking to explore AI to do a safety management system for a commercial vessel, but perhaps by the time I entered the relative info it might be as easy to do the job myself, as I it's not a particularly complex operation I'm considering.
I am into old British sports cars and do a lot of work with club members helping them sort out their car's issues.
I googled servicing the brakes on a car. The advice that ChatGPT AI threw up was dangerous rubbish. AI has a long way to go if people with little expertise are to rely on it.
Some handy rules for Chagpt:
"Don't make assumptions. Only state what is directly known or cited"
"Be literal. Do not extrapolate beyond the data provided."
If you need a bunch of rules, it's easier to put them in a text file, upload it and ask Chagpt to reference it when answering questions.
This is an example of tying some rules together, this is one of the 20 x odd files we enforce for when we are using Chagpt to do a gap analysis on a submission for a government tender.
First time post from and old sailor who's getting back into it after 15 years, but i've been into AI since the 90's and making a custom AI/LLM specifically for sailing would be a very interesting thing to do. ChatGPT is fun, but seems to have a very generic data set for the hobby.
The other thing I thought was interesting on the AI side was frigate for object detection (frigate's an open-source network video recorder with AI for detecting thing's), but one re-trained for the top of the mast looking out over the ocean would be a very interesting project.
