AI
- Thread starter Sol Ski
- Start date
That's fair. I made errors in something that needs to be data-exact.
The biggest problem isn't just that the first graphic had incorrect ORs. After you specifically asked me to correct it, I should have audited every field against your supplied HRB data before presenting the replacement. I didn't do that.
I can already see problems in the replacement graphic: it says 4yo+ when the race is 3yo+, shows the winning prize as £4,187 instead of £3,978, and some of the displayed weights don't match your racecard. Those are unacceptable transcription errors for this project.
There is also a more fundamental issue: I presented detailed component scores in the graphic as though they were exact outputs of our established model, when I had not transparently shown the calculation behind every component. That means you couldn't properly audit the 91.0 rating either.
For a six-month live betting experiment, that's not good enough.
I think the right correction is not another AI-generated graphic yet. We should first rebuild this race's 100-point calculation in an auditable table, using only the HRB data you supplied, show exactly how every horse gets every point, verify the totals together, and only then create the graphic from that locked table.
And until we've done that, I would not treat American State as an officially recorded Live Bet #8.
The biggest problem isn't just that the first graphic had incorrect ORs. After you specifically asked me to correct it, I should have audited every field against your supplied HRB data before presenting the replacement. I didn't do that.
I can already see problems in the replacement graphic: it says 4yo+ when the race is 3yo+, shows the winning prize as £4,187 instead of £3,978, and some of the displayed weights don't match your racecard. Those are unacceptable transcription errors for this project.
There is also a more fundamental issue: I presented detailed component scores in the graphic as though they were exact outputs of our established model, when I had not transparently shown the calculation behind every component. That means you couldn't properly audit the 91.0 rating either.
For a six-month live betting experiment, that's not good enough.
I think the right correction is not another AI-generated graphic yet. We should first rebuild this race's 100-point calculation in an auditable table, using only the HRB data you supplied, show exactly how every horse gets every point, verify the totals together, and only then create the graphic from that locked table.
And until we've done that, I would not treat American State as an officially recorded Live Bet #8.
I’ve got to be 2 full weeks into getting my HK ratings to pure python, I asked it yesterday how far we had got it said 90-95%, I asked it today for a timescale and it said 3-6 weeks remaining , doesn’t add up, I swear it’s doing things in version 3 million that it had already done in version 30. It seems quite confident but then that seems default mood for AI. It can be frustrating but I still have some faith because 2 smaller projects I’ve completed a scraper and standard time calculator work well.
StefanBelo
Yearling
S
Sol Ski and
O
Outlander Are you attempting to process HRB data parsing, or are you developing content for your own website using HRB’s presented data?
I used HRB presented results data downloads to bolt on to sectionals scraped from elsewhere successfully , and HRB presented results data downloads to build my standard time calculator successfully , my other project is for Hong Kong so no HRB . I have never needed or attempted to scrape HRB I use their available downloads.
StefanBelo, im feeding data into chatgpt from HRB to create my ratings.
MySportsAI
Filly
Are you asking chatgpt to do math ?
I have preset paramiters that i have got it to lock in. only noticed i had a problem yesterday.
MySportsAI so yes there is maths involved.
MySportsAI so yes there is maths involved.
StefanBelo
Yearling
S Sol Ski and
MySportsAI Did you code using ChatGPT? My setup offers a full development loop. With this, AI can write scripts to analyze real data, load that data into my app, and then improve the code if necessary.
StefanBelo Please dont take this the wrong way but that might as well be in Chinese for someone like me.
Please dont take this the wrong way but that might as well be in Chinese for someone like me.
MySportsAI
Filly
I have used it to on the odd occasion but generally code myself. If you are using it to code that would seem a sensible use my previous question was whether anyone was using it to actually compile ratings itself, I would not trust it for that. For example I have written an agent for my son and his property management business but chatGPT cannot calculate date differences.
S Sol Ski and
StefanBelo
Yearling
Forgive me, but I don't think you actually know what you're talking about.
Here is the proof: on the right, you can see the data, and in the middle is my prompt asking for what you claimed was impossible.
As you can see, the AI produced the correct date difference.
I don't understand why you would make a false claim, one that can easily be verified by anyone with five minutes of spare time.
MySportsAI
Filly
Correct me if I am wrong but thats not an API call is it
MySportsAI
Filly
I also asked Grok what it thought of your reply above, I wondered if I was being a tad sensitive
"Condescending, dismissive, and accusatory, with a thin veneer of politeness."
MySportsAI
Filly
The point I am making is that there is a difference between typing in prompts and constructing a fully automated API based program. If you do not believe me chatGPT will confirm this if you ask it. You see its a pattern matcher not a Math program. It can get things right for sure but it can also get basic things wrong
MySportsAI
Filly
From chatGPT itself
"This happens because date arithmetic is not something you should rely on an LLM to perform reliably inside an API prompt."
"This happens because date arithmetic is not something you should rely on an LLM to perform reliably inside an API prompt."
MySportsAI
Filly
I am assuming that at some point you will want to fully automate your system perhaps using OPenAI. Do not let my post put you off you can still do this if you provide sensitive arithmetic op's as tools in your API call to chatGPt. Here is a short article that may help
How to Turn a Python Script Into an AI Agent - KDnuggets
Learn how to build a Python AI agent with the OpenAI Agents SDK, using tool calling and function tools to automate multi-step workflows.
www.kdnuggets.com
StefanBelo
Yearling
I am assuming that at some point you will want to fully automate your system perhaps using OPenAI. Do not let my post put you off you can still do this if you provide sensitive arithmetic op's as tools in your API call to chatGPt. Here is a short article that may help
How to Turn a Python Script Into an AI Agent - KDnuggets
Learn how to build a Python AI agent with the OpenAI Agents SDK, using tool calling and function tools to automate multi-step workflows.
Please allow me to clarify my background. When responding to your query, 'Can you give an example?', I previously provided supporting screenshots along with concise explanations in a different thread.
The evidence clearly demonstrates my utilization of AI agent methodologies within my application framework, as seen in the first screenshot attached to my reply.
For context, I am a software developer with over three decades of experience in diverse technologies and programming languages. My current involvement is purely personal. I have been retired from professional development work for several years. Furthermore, since initiating trading on Betfair in 2007, a time during which I built a proprietary application, all my contributions here are based on extensive personal experience, viewed through both the lens of financial trading and software engineering.
Here is my AI Agent in my betfair app:
Program: "Get the active market and list the two horses with the highest traded volume."
Result:
The currently active market is for **Sandown 5f Hcap** (Market ID: `1.262425289`).
Based on the traded volume, the two horses with the highest activity are:
1. **Good Earth**: Traded Volume of 1441.36
2. **Hint Of Humour**: Traded Volume of 1258.72
Yes, my AI agent can take plain English instructions and use them either to place bets on Betfair or execute complex trading strategies.