• Hi Guest Just in case you were not aware I wanted to highlight that you can now get a free 7 day trial of Horseracebase here.
    We have a lot of members who are existing users of Horseracebase so help is always available if needed, as well as dedicated section of the fourm here.
    Best Wishes
    AR

Artificial intelligence (AI) Large Language Models (LLMs)

Paul Brown

Paul Brown

Gelding
One thing I've found with 'AI' is that it can be very frustrating. If you train a dog to sit, you must give it the command each time. It's the same with AI you need to give it the same clear instructions every time, otherwise it starts to stray.
For example, say you wanted the PRB (Percentage of Rivals Beaten) for a race. My command would be:
"Give me the PRB (Percentage of Rivals Beaten squared) for every horse in the race and keep them in racecard order. Also include a 'Copy to clipboard' button so I can just press it instead of having to select all the text. "I would get a racecard (for example from 'The Daily Racer'), copy the form, and paste it into Notepad first to convert it to plain text.
Then I would paste it into Grok (or another AI) and get the result. Before feeding any race into Grok or any other AI, I always give it the same instructions. I've found that if you don't, it can easily veer off track. Large Language Models (LLMs) can also hallucinate. Just today I asked an AI to fetch me the prices for a racecard. It did this very quickly.
When I asked where it had got the prices from, the answer was;
"I hold my hand up, couldn't scrape a website so I made them up." Definition of AI hallucinations: AI hallucinations occur when a large language model generates false, misleading, or completely fabricated information that sounds highly confident and plausible.
 
what drives me nuts is i have the simplest of task for AI to help me with and i get it running perfect then out the blue he says i cant help you bet so you have to explain its not bets its following my instructions so he goes happy again for while keeping saying cant give you tips i dont want tip just tell me what comes out best from my info then were happy again then he says days later i cant say if you should bet in this race or not i am like well don't say just do as i ask worse than a wain he is. but the other side of it is your typing out your thoughts and it does help you to analyse in writing and deff better than trying to remember things just look it up again.
 
gerry said:
what drives me nuts is i have the simplest of task for AI to help me with and i get it running perfect then out the blue he says i cant help you bet so you have to explain its not bets its following my instructions so he goes happy again for while keeping saying cant give you tips i dont want tip just tell me what comes out best from my info then were happy again then he says days later i cant say if you should bet in this race or not i am like well don't say just do as i ask worse than a wain he is. but the other side of it is your typing out your thoughts and it does help you to analyse in writing and deff better than trying to remember things just look it up again.
Click to expand...
What model are you using, gerry gerry ?
 
Paul Brown said:
One thing I've found with 'AI' is that it can be very frustrating. If you train a dog to sit, you must give it the command each time. It's the same with AI you need to give it the same clear instructions every time, otherwise it starts to stray.
For example, say you wanted the PRB (Percentage of Rivals Beaten) for a race. My command would be:
"Give me the PRB (Percentage of Rivals Beaten squared) for every horse in the race and keep them in racecard order. Also include a 'Copy to clipboard' button so I can just press it instead of having to select all the text. "I would get a racecard (for example from 'The Daily Racer'), copy the form, and paste it into Notepad first to convert it to plain text.
Then I would paste it into Grok (or another AI) and get the result. Before feeding any race into Grok or any other AI, I always give it the same instructions. I've found that if you don't, it can easily veer off track. Large Language Models (LLMs) can also hallucinate. Just today I asked an AI to fetch me the prices for a racecard. It did this very quickly.
When I asked where it had got the prices from, the answer was;
"I hold my hand up, couldn't scrape a website so I made them up." Definition of AI hallucinations: AI hallucinations occur when a large language model generates false, misleading, or completely fabricated information that sounds highly confident and plausible.
Click to expand...
According to this guy most of us use AI incorrectly. I find his videos interesting and am starting to learn the correct way to ask questions and prompting -
 
Spireite said:
According to this guy most of us use AI incorrectly. I find his videos interesting and am starting to learn the correct way to ask questions and prompting -
Click to expand...
This video is exactly what I have been saying. AI in one form or another is here to stay but be guarded because its only as good as you are.
 
gerry said:
i am just basic copilot on pc nothing clever.
Click to expand...
Gerry I think copilot is good enough for some things. AI is changing all the time and will in most cases be improved. I, like a lot of older people find it daunting to learn new things. One thing that came through when I checked my DNA was that I have a lot of determination. How it arrived at this is anyones guess. I would have liked it to add that I was highly intelegent but it didn't mention this. :cool: I am what I am. Paul
 
Spireite said:
According to this guy most of us use AI incorrectly. I find his videos interesting and am starting to learn the correct way to ask questions and prompting -
Click to expand...
I started using GPT almost the moment it became available in the UK (just before COVID) to analyse UK racing form. Things are different now from the mainly US racing focus/efforts of the early GPT developers. Most of the points in the video you link to are valid, especially not "over-stuff" GPT with duplicate data and avoiding automated scraper/web search data input. The V15 Early Doors blog has developed into a fascinating hobby and my legacy record of human and machine working together. At its basic form, I input the data, 'obby does the backend work, and I decide what or if to bet on.
 
AJ the Hobbyist said:
I started using GPT almost the moment it became available in the UK (just before COVID) to analyse UK racing form. Things are different now from the mainly US racing focus/efforts of the early GPT developers. Most of the points in the video you link to are valid, especially not "over-stuff" GPT with duplicate data and avoiding automated scraper/web search data input. The V15 Early Doors blog has developed into a fascinating hobby and my legacy record of human and machine working together. At its basic form, I input the data, 'obby does the backend work, and I decide what or if to bet on.
Click to expand...
Exactly - You decide on what or if to bet on - there's too many people who are followers and blindly bet, whether it be so called professionals posts in the paper media or the numerous dodgy social media posters.

I have many followers on X, I post selections and stress they're selections but the number of new followers who ask, how much do I bet on the selections, do I recommend singles, doubles, lucky 15's etc......My reply is always, these are selections, take a look and make your own mind up.
 
i had big fall out with ai last night spent 2 hours going over all this afternoon racing as golfing this morning and when i asked it to go back to epsom the first meeting i did it got it all mixed up. i thought it was smart but said it cant go back that far i said we just done it 1 hour ago my god it tried to reason why it was forgetful but i had enought so told it to get lost it was not happy my point was i am giving it good information to store and it should be appreciative to gain all it can as it is AI after all but i will do it my self now had enough of it not doing exsactly what i asked if it was secretary would be sacked on spot.
 
You can provide your LLM with tools that give it access to data, which allows you to ask anything you want. If data is available, hallucinations during the answering process are minimized. You can see this in my screenshot for the question: "Retrieve favorite horse racing events in GB and IE, and report on which racecourses are hosting today's races." I received a relevant answer; as shown in the screenshot, the LLM did some thinking before calling a tool for data, then performed further processing before calling another tool for specific details.

1780651261151.png
 
gerry said:
i had big fall out with ai last night spent 2 hours going over all this afternoon racing as golfing this morning and when i asked it to go back to epsom the first meeting i did it got it all mixed up. i thought it was smart but said it cant go back that far i said we just done it 1 hour ago my god it tried to reason why it was forgetful but i had enought so told it to get lost it was not happy my point was i am giving it good information to store and it should be appreciative to gain all it can as it is AI after all but i will do it my self now had enough of it not doing exsactly what i asked if it was secretary would be sacked on spot.
Click to expand...
Exactly Gerry. If you don't repeat the instructions every time, it will go off track. Save your instructions to microsoft word or to notepad or write them in every time. Consider copilot or any other to be a small child who needs constant instruction. Liken it to a golfer who needs to go through the same procedure every time in order to keep their ball on the fairway. If you don't have Microsoft word use Google Keep.
 
Last edited:
You must log in or register to reply here.
Back
Top