The trick to making good content driven by LLMs is threefold:
1. Using the right LLM model. Since it would be important in this case to have more up to date information, you should make sure you utilize the "Deep Research" version of your model so that the provider actually crawls the web right now for the latest info instead of using potentially older training data.
2. Add a lot of detailed context and parameters to your prompt. It's not enough to just simply say give me the latest on all Chelsea loanees. What I would do is first get a list of the loanees that you are sure is accurate. Then give the LLM that list and tell it exactly what you want it to do for each player and what format you want it to put it in. Like, for example "for each player get stats x, y, and z with their loan teams and give detailed breakdowns of their games from the past two weeks that includes a, b, and c, etc."
3. Finally, no matter how good you prompt the AI and no matter how good the model is, you can't just publish whatever it generates. I've found that even when it does a good job, there is a lot of value to spending 10 minutes at least (i.e. it can be a short amount of time) going through and tweaking the output content in order to smooth out rough edges and add your own thoughts as appropriate.
I completely respect your decision to want to free some time up but this just indicates Ai can’t do what you can do. What ever you decide to do is fine with me.
I voted the AI amended because I’m not yet ready to give up on the concept. However I was very nearly in the “over my dead body” section but am willing to see how it can be used but the difference was stark I’m afraid!
I used to read the loan army on each player religiously as I found it informative, well presented and made me aware of what was happening with each player and I can well believe it was time consuming to compile.
However I read 4 or 5 AI reports and the information just got less not more for example is Ben Chilwell injured? Again?
Luke this is one of my favourite monthly articles as I’ve mentioned to you in the past. On the poll I chose AI with your input and now having read the article as a whole I think your personal input is essential.
I fully understand that you’re trying to free up time and respect that. It’s your personal knowledge that I think makes the article such a good read.
Could I also suggest that at this time of the season would be worthwhile you adding, particularly for the newly recruited and younger players, what you would expect for them for the following season. i.e. another loan at a higher level, a sale, a first team squad position, et cetera.
Thanks Craige, think I am going to do a mixture going forward, get AI to give me some info but go through and research each player myself as before to see what they've been up to!
Not sure if you remember back to last year but I did an overall roundup article of each players season, gave them a rating and then my verdict of what should happen going forward. I will do that agin come the end of the season!
Hey Luke loved the more laborsome outputs, but maybe a bridging solution could be asking AI to leverage Sofascore and G/A plus key stat from last 3 games or so? And asking AI to include development of time played (like Carney is a black box in this format, but has actually gotten much better throughout his stint). Hope that helps?
I felt that the AI version is missing information sentiment and color around the loan players performances and additional information about whether they’ll come back or be transferred.
I’m also surprised that for some players it says no match data after March 5. That doesn’t seem correct but I may be wrong.
I like the new format
The trick to making good content driven by LLMs is threefold:
1. Using the right LLM model. Since it would be important in this case to have more up to date information, you should make sure you utilize the "Deep Research" version of your model so that the provider actually crawls the web right now for the latest info instead of using potentially older training data.
2. Add a lot of detailed context and parameters to your prompt. It's not enough to just simply say give me the latest on all Chelsea loanees. What I would do is first get a list of the loanees that you are sure is accurate. Then give the LLM that list and tell it exactly what you want it to do for each player and what format you want it to put it in. Like, for example "for each player get stats x, y, and z with their loan teams and give detailed breakdowns of their games from the past two weeks that includes a, b, and c, etc."
3. Finally, no matter how good you prompt the AI and no matter how good the model is, you can't just publish whatever it generates. I've found that even when it does a good job, there is a lot of value to spending 10 minutes at least (i.e. it can be a short amount of time) going through and tweaking the output content in order to smooth out rough edges and add your own thoughts as appropriate.
I completely respect your decision to want to free some time up but this just indicates Ai can’t do what you can do. What ever you decide to do is fine with me.
I voted the AI amended because I’m not yet ready to give up on the concept. However I was very nearly in the “over my dead body” section but am willing to see how it can be used but the difference was stark I’m afraid!
I used to read the loan army on each player religiously as I found it informative, well presented and made me aware of what was happening with each player and I can well believe it was time consuming to compile.
However I read 4 or 5 AI reports and the information just got less not more for example is Ben Chilwell injured? Again?
Keep the good work up.
Luke this is one of my favourite monthly articles as I’ve mentioned to you in the past. On the poll I chose AI with your input and now having read the article as a whole I think your personal input is essential.
I fully understand that you’re trying to free up time and respect that. It’s your personal knowledge that I think makes the article such a good read.
Could I also suggest that at this time of the season would be worthwhile you adding, particularly for the newly recruited and younger players, what you would expect for them for the following season. i.e. another loan at a higher level, a sale, a first team squad position, et cetera.
Thanks from all of us for your time and effort
Thanks Craige, think I am going to do a mixture going forward, get AI to give me some info but go through and research each player myself as before to see what they've been up to!
Not sure if you remember back to last year but I did an overall roundup article of each players season, gave them a rating and then my verdict of what should happen going forward. I will do that agin come the end of the season!
Perfect and thanks again
Hey Luke loved the more laborsome outputs, but maybe a bridging solution could be asking AI to leverage Sofascore and G/A plus key stat from last 3 games or so? And asking AI to include development of time played (like Carney is a black box in this format, but has actually gotten much better throughout his stint). Hope that helps?
I felt that the AI version is missing information sentiment and color around the loan players performances and additional information about whether they’ll come back or be transferred.
I’m also surprised that for some players it says no match data after March 5. That doesn’t seem correct but I may be wrong.
I really enjoyed the Loan Army segment, but feel that this AI version is wide of the mark.
Personally, I would quite like just the stats, either rolling from start of loan, or for the club's last 3 games.
- minutes played.
- appearances/subs.
- goal contributions/goals conceded (for defenders).
- average player rating.
- team results.
Something like that. I am not sure if AI can provide that. I am not against AI, but can't see a lot of what was written changing much week after week.
Thanks Joe, really appreciate feedback!
At the end of the day my goal is to provide you all with what you want to see in an easy to read/see article!