Chat GPT & Generative AI tools
-
ShaunWhite
- Posts: 10388
- Joined: Sat Sep 03, 2016 3:42 am
I had a very long programming session on Chat yesterday and it wasn't working very well. It took about a dozen iterations to get a simple geometric calculation right - each offering I rejected and showed proof the numbers generated were wrong.
Near the end it seemed to lose its cool - it came up with headings and comments like this which I'd never seen it do before :
So with this sort of response you can't help but wonder who is behind this Chat or is it sentient - so I asked it about emotion - the responses are below - they were all automatic - the response was instantaneous and the display speed outdid any manual typing ... I would give that a high mark on the Turing test - the future may be nearer than we think !
Near the end it seemed to lose its cool - it came up with headings and comments like this which I'd never seen it do before :
So with this sort of response you can't help but wonder who is behind this Chat or is it sentient - so I asked it about emotion - the responses are below - they were all automatic - the response was instantaneous and the display speed outdid any manual typing ... I would give that a high mark on the Turing test - the future may be nearer than we think !
You do not have the required permissions to view the files attached to this post.
-
sionascaig
- Posts: 1611
- Joined: Fri Nov 20, 2015 9:38 am
-
sionascaig
- Posts: 1611
- Joined: Fri Nov 20, 2015 9:38 am
Yes it stood out since I'd never encountered it behaving like that before and I use it a lotsionascaig wrote: ↑Sat Feb 01, 2025 11:08 amDo you not find it a bit weird that it talks like that though?
(its not your pal !)
default 4o for that thread - usually too lazy to switch models
Not sure traits existed when I signed up - none ticked anyway but has the instruction to give short responses and no explanation unless asked - seems to ignore that more than it used to these days. It did offer a new model this morning so it is still an evolving "being"
I've been using AI to build a tennis forecasting model with some moderate success. Things i've discovered so far...
Ai is still young in it's development and you can't 100% rely on it
It hallucinates, it gave me complete list of todays bets with forecasts and none of them existed.
You have to be very specific in your prompt instructions
You have to have a clear plan as to what you want it to do before asking
Deepseek is often too busy and takes forever to work
If you want start using it for betting I have found a simple prompt giving it some bets to forecast is a good start. Make sure it's connected to the internet and ask for estimated win% and estimated odds. Tell it not to refer to the bookies sites, then you can get odds that you can seek value bets from. Once you have it forecasting you can start refining the process.
The measure of success will lie in the backtesting you get it to do and the level of accuracy it returns, This metric can be refined over time by expanding it's range of research e.g. check weather conditions, travel time to match, injuries, etc.
Just a few hints for anyone wanting to get into it more seriously. I'm getting around 67% hit rate currently and am tweaking to slowly improve the rate. It's really a whole new world and will bring huge profits to those who get their head around it.
Good luck
Brian
P.S. Agentic AI is the new kid on the block and has only been really useful for a couple of months. It allows you to build agents to perform specific tasks like to go off and research current injuries for players. You can set it to do it regularly to support your betting strategy.
Ai is still young in it's development and you can't 100% rely on it
It hallucinates, it gave me complete list of todays bets with forecasts and none of them existed.
You have to be very specific in your prompt instructions
You have to have a clear plan as to what you want it to do before asking
Deepseek is often too busy and takes forever to work
If you want start using it for betting I have found a simple prompt giving it some bets to forecast is a good start. Make sure it's connected to the internet and ask for estimated win% and estimated odds. Tell it not to refer to the bookies sites, then you can get odds that you can seek value bets from. Once you have it forecasting you can start refining the process.
The measure of success will lie in the backtesting you get it to do and the level of accuracy it returns, This metric can be refined over time by expanding it's range of research e.g. check weather conditions, travel time to match, injuries, etc.
Just a few hints for anyone wanting to get into it more seriously. I'm getting around 67% hit rate currently and am tweaking to slowly improve the rate. It's really a whole new world and will bring huge profits to those who get their head around it.
Good luck
Brian
P.S. Agentic AI is the new kid on the block and has only been really useful for a couple of months. It allows you to build agents to perform specific tasks like to go off and research current injuries for players. You can set it to do it regularly to support your betting strategy.
I've never seen it talk like that to me. Though I do use custom instructions to shape the response.sionascaig wrote: ↑Sat Feb 01, 2025 11:08 amDo you not find it a bit weird that it talks like that though?
(its not your pal !)
The biggest mistake people are making at the moment is using large LANGUAGE models to do statistics.betghost wrote: ↑Sat Feb 01, 2025 12:48 pmI've been using AI to build a tennis forecasting model with some moderate success. Things i've discovered so far...
Ai is still young in it's development and you can't 100% rely on it
It hallucinates, it gave me complete list of todays bets with forecasts and none of them existed.
You have to be very specific in your prompt instructions
You have to have a clear plan as to what you want it to do before asking
Deepseek is often too busy and takes forever to work
If you want start using it for betting I have found a simple prompt giving it some bets to forecast is a good start. Make sure it's connected to the internet and ask for estimated win% and estimated odds. Tell it not to refer to the bookies sites, then you can get odds that you can seek value bets from. Once you have it forecasting you can start refining the process.
The measure of success will lie in the backtesting you get it to do and the level of accuracy it returns, This metric can be refined over time by expanding it's range of research e.g. check weather conditions, travel time to match, injuries, etc.
Just a few hints for anyone wanting to get into it more seriously. I'm getting around 67% hit rate currently and am tweaking to slowly improve the rate. It's really a whole new world and will bring huge profits to those who get their head around it.
Good luck
Brian
P.S. Agentic AI is the new kid on the block and has only been really useful for a couple of months. It allows you to build agents to perform specific tasks like to go off and research current injuries for players. You can set it to do it regularly to support your betting strategy.
67% is around the random base rate for a favourite in a Tennis match, I think.
-
ShaunWhite
- Posts: 10388
- Joined: Sat Sep 03, 2016 3:42 am
Yes they haven't understood how to use the tool and expect too much. If people just ask for stats based on public knowledge it's going to be flaky, but you can upload files of up to 50mb and ask for whatever analysis you want and it's great. Or if it's struggling it will write the code /excel or whatever for you to continue it yourself.
It's not just gigo it's about learning how to work within it's limitations too.