Mars_Blackmon
-
Posts
1,992 -
Joined
-
Last visited
Content Type
Forums
Articles
FAQs
Online Manual
Support: Blocks
Support: Games
Bug Tracker
SIGames Manual (beta)
Profiles
Posts posted by Mars_Blackmon
-
-
46 minutes ago, The3points said:
I still don't think you get the point that has been communicated for the entirety of this forum. The tactic, in my opinion, is subpar: the wings are exposed, attacking play one-dimensional.
"Know the game" means knowing exactly what each click does to the game engine. "Know the game" is understanding what factors change when you restart the game. "Know the game" means having 100% access to the formulas that tempo, and directness change, when these forums have three different definitions for each. Right know, ChatGPT "knows the game" as much as an intermediate player who gets the words and what they mean but not much of the significance and how they act in-game.
This is what I see but anyway no matter, this post this tactic on a forum anywhere, reddit, here wherever, they'll say that it is too attacking, contradicting styles, etc. For example many would ask why don't you have sweeper keeper for high line, that the wing is too aggressive, and so on and so forth. Especially in the original tactic, which while you say it's a good base (and it is, the instructions work quite well), how many new FM players are going to change 3 roles? My point is treating chatGPT like a trusty resource is worse than using Wikipedia for an academic article: chatGPT realistically doesn't know what's it's talking about. If it was hooked up with how the game worked, the "secret formula", then it would be trusty because it does know what it's talking about
I'm not sure what you didn't include. Nonetheless, we can go all day about the best tactics to use and what each code is programmed to do in the game. The same conversations are happenngi on multiple FM forums about tactics people manually develop. The thing that matters is that it worked in the game. It won the league with a team that was predicted to be 4th. I am not looking for the next underdog tactic that takes a bottom team to #1. I am pointing out that it can put together a coherent tactic even more than the preset tactics do, which does not score well using guidetofm and ratemytactic calculations if those hold any weight. It's a usable tactic and not put together randomly.
That's the bottom line.
If I am a user who isn't good at developing tactics, that's an adult. Would I instead spend all day on forums waiting for a reply, get a response only to get told to use the search function, or value my time and get that information in minutes? I value my time, so I would rather spend those wasted hours enjoying the game.
But to each thier own.
0 -
41 minutes ago, phnompenhandy said:
Here's me at the start of my save. I give myself the worst possible attributes and personality. For the first few months, players don't listen to you in team talks, and don't like your training. But as long as you get adequate results, the effects don't last long at all. The only lingering effect as far as I can tell is that the board never meet your requests.
Yeah, attributes have an effect on the things you mentioned, but I am speaking on characteristics. I would hope that these are more than visual and have some benefits & risks associated with them. For example, not being loyal to players should make it harder to attract them to the club if they know you operate with a high turnover rate.
0 -
22 minutes ago, The3points said:
Original tactic had two WBs instead of FBs.
But the last line is pointless: that's not what this is about. I'm not you looking for a tactic, but anyone using chatGPT would be, and the tactic they would get is flawed. Simple as that. Now if it knew about the game, was integrated with the game's logic etc. it would be a game-changer: but for now it's pretty flawed and not that close to improving the status quo of FM
Error on my part, but it's still 5*
You can recreate it with my few edits to confirm if that matters.
And what do you mean by "Know about the game" It has data on the game up until 2021, and if you use the web browser version, it can retrieve data from the internet. We aren't in the phase of AI thinking for itself. Is that the disconnect here? Machine learning is the first phase of AI. Were are years if not decades from the AI you see in movies at a commercial level...
0 -
3 minutes ago, themadsheep2001 said:
I dont think you know what a straw man argument is. The context here is whether GPT can offer working well thought out tactics from prompts. It cannot, thus it fails at being a useful tool for FM. You can echo all you like, but its been repeatedly demonstrated it offers up flawed tactics.
Do you know people are getting paid lots of money for prompt engineering? You can't prompt right. It is a personal problem. Many people use the tool for FM without any problems, and others use it for way more complex things than video games.
0 -
\
1 minute ago, themadsheep2001 said:I dont think you know what a straw man argument is. The context here is whether GPT can offer working well thought out tactics from prompts. It cannot, thus it fails at being a useful tool for FM. You can echo all you like, but its been repeatedly demonstrated it offers up flawed tactics.
You are arguing a point that I have not made. Your man problem is that it doesn't know FM or think for itself. You made that point multiple times in here, and no, but you mentioned that.
0 -
7 minutes ago, themadsheep2001 said:
"It doesnt think or come up with its own ideas" - thats precisely the problem.
"The concept of prompting is to get clear, relevant information back" again it doesn't give back clear relevant information. Thus it fails at its task here
I'm not the only person to have pointed it out, but you seem intent on ignoring this so i will spell it out clearly.
It cannot advise clearly, since the tactics it offers up are inherently flawed. Because it cannot actually properly filter out good information from bad well enough. And since it has no connection to the game it has no true context to place its data around. In terms of FM usage it is still rubbish. Thankfully SI haven't put it anywhere near FM and thats a good thing.
Again, straw man arguments.
And the examples I posted here weren't exactly prompted right, which I have pointed out multiple times. It doesn't need to know the game because the data it gets is from sources that played it. A good prompt adds context. I echo that you don't know what you are talking about.
0 -
7 hours ago, themadsheep2001 said:
There are no better prompts, the software isnt smart enough, doesn't sort through whether information it dregdes up is good enough and doesn't understand FM in any context. Nothing has changed since the topic was started in this regards.
Great insight. Your choice of words shows that you have no idea what you are talking about. The concept of prompting is to get clear, relevant information back. "Smart enough." It doesn't think or come up with its own ideas, and for some reason, you keep bringing up the point. You're trying to make a straw man argument. It functions based on patterns learned from the data it was trained on or now with GPT4, the internet. It filters through the data it has learned from or retrieved. It predicts the next word based on the input prompt, trying to generate coherent and contextually relevant text based on the parameters of the prompts, which is similar to coding but in natural language.
0 -
6 hours ago, themadsheep2001 said:
Exactly. If people actually want this to be good, they need to accept where its rubbish so it can improve, not stick tneir fingers in their ears.
In terms of FM, its still not good enough to be anywhere near the game.
Yet you continue to add nothing to the thread but give an opinion on how wrong someone is instead of trying to come up with better prompts.
0 -
The question i have is what effects does this have on gameplay? Seems like this along with our characteristics is just visual.
0 -
1 hour ago, The3points said:
Sure, but at that point you've changed almost half the tactic: 2 FBs, 1 CB, 1 CAM, 3 instructions. Realistically chatGPT isn't the resource of choice you would use for this sort of thing,
I changed 1 BPD to a CD (Even though 2 BPD works well in the game, ratemytatcic doesn't like it), and I changed the CAM from attacking to support. That's hardly half the tactic. the offside trap works, but I was suggested to use it situationally. Same with counter and counter press. But we all know that pretty much works with any high-press tactic.
Most importantly, no one is forcing you to use it.
0 -
2 hours ago, enigmatic said:
I agree with that much
It's just a tool that's supposed to give you tactical advice that's mostly wrong unless you're advising it isn't a particularly useful tool!
I'm not going to argue with your opinion. Especially since you are ignoring the concept of input = output.
0 -
41 minutes ago, enigmatic said:
Neither do LLMs
But even an LLM knows to distinguish between accurate and false information when prompted...
At the end of the day, it's a tool. and with tools and machines, it's always the person behind it, not the tools itself.
0 -
15 minutes ago, enigmatic said:
More just because you're getting nuggets doesn't mean there was ever any chicken involved
Nuggets don’t have bones
0 -
4 minutes ago, enigmatic said:
tbh it looks like a pretty standard gegenpress you could find in these forums, or any third-party site (in downloadable form) with added nonsense related to your statistics (nothing wrong with mixed set piece delivery, but the average aggression rating of the team has nothing to do with whether it's the right option or not).
Suggesting the team lacks eccentricity and that you should use an offside trap in possession gives you an idea of how good the tactical insight actually is.
When SI's tooling gives crap recommendations like that I file bug reports, and whilst they don't always fix the bugs at least they actually can...
Eat the meat, spit out the bones.
0 -
2 hours ago, The3points said:
These were the results I got.
Then after BPD adjustment.
Never mind the fact that symmetrical roles in tactics isn't the best
This is the actual tactic. I forgot that I also changed the AP from attack to support. Counter press and Counter were removed as GPT mentioned that they should be situational after I raised the conflicting error. It's the same with the offside trap after I told it about the updated strategies for the offside trap in the game.
Also, this is without going deep into each player and their best roles. So, a CF or even a BPD at the lower league most likely wouldn't have been suggested if I got even more detailed.
I loaded an Excel file and told it to show strengths and weaknesses. (You have to ask GPT to exclude the GK as it will factor their attributes in the overall average and vice versa for the GK attributes for the team; You could go deeper than that and separate each position like defender, midfield, and attack.) Then, ask it to create a suitable tactic. Then, we fixed any errors.
This is data-driven. The better data and prompts you put in, the better results. People seem to expect GPT to have its own AI brain and are quick to point out any errors....
0 -
8 hours ago, srvngrc said:
I don't know if this idea has been discussed before. If yes please accept my apologies.
As you know, we play against AI in FM and things can get easy after a while. AI managers can't be competitive enough. But what if we played the game against real opponents, wouldn't it be more fun and challenging? I have an idea of what it could be like. Millions of real players are making individual careers in FM simultaneously. But to make multiplayer online careers, everyone has to take their time and be online at the same time, and it's not easy.
But can't AI imitate other FM players? I think it's not too hard to set. For example, when you start a career with Arsenal, AI can copy the tactics and transfer moves of FM players who have had successful City and Liverpool careers elsewhere in the world. Thus, a kind of offline multiplayer career will be possible.
Some of you may like this idea or some of you may seem ridiculous. I'd love to hear your thoughts.Machine learning is a subset of AI, and considering this has been somewhat done with games before, it is not out of the realm of possibility. However, if this would ever become a thing with games, then I would probably be done because I stay away from online games, especially with sports, because of the "unrealistic" aspect that comes with it.
LOL, if the AI in FM already spam 4-2-3-1, I can imagine playing against the AI that all spam the same OP tactics.
Learning from a user base can still create undesirable results. However, if we scrum the data from actual Football games and transfer patterns, then we are talking!
Ironically, this is somewhat similar to what they are doing with kinematics and motion matching for animation.
0 -
52 minutes ago, The3points said:
I think the tactic is quite poor to be honest. Ignoring the errors such as the winger being in left and right midfield, the duplication of offside trap, higher line of engagement for FM23 etc. the tactic just doesn't work that well. The tactic has almost the exact same qualities up the right as up the left, 3 out of the front four are on attack, both fullbacks move forward and this is all with Oxford City, a National League South team. While more "personalised" than the preset, I think it's not the best tactic for general use.
I didn't share the entire chat. The wingers are attacking midfielders, hence, the formation and the offside trap duplication were fixed. The adjustments I made were only using one BPD. Other than that, not only did we win the league during a simulation to test the tactic, but the tactic scored five stars on ratemytactic after the adjustment with the BPD.
Ratemytactic is another tool that can help with tactics, but that was used to make sure that penetration, support, and solidity all scored well. So it did work well logically.0 -
The paid version of GPT4 has added a few tools since April. One that stands out is the data analysis tool. Using the tool, I gave it an Excel file, and asked it to create a tactic suitable for the team. I made some adjustments but definitely a good starting point and more personalized to your team vs. the preset tactics.
QuoteGreat, we've successfully loaded the data this time! The dataset is quite comprehensive and includes various attributes for each player on the Oxford City team, based on Football Manager 2023 data.
Here are some of the key columns:
- **Position Selected**: The position the player is primarily chosen for.
- **Name**: Player's name.
- **Age**: Player's age.
- **Nat**: Nationality.
- **Club**: The club they play for (Oxford City in this case).
- **On Loan From**: The club the player is loaned from, if applicable.
- **Position**: The player's primary position.
- **Sec. Position**: The player's secondary position.
- **Player Status**: Any special status for the player like 'Wnt - Rst' (Wanted - Rest).
- **Preferred Foot**: The foot the player prefers to use.
- **Height, Weight**: Physical attributes.
- **Various Skill Ratings (e.g., Acc, Aer, Agg)**: Ratings for different skills and attributes.
- **Wage, Value**: Financial details.
- **Transfer Status, Asking Price**: Transfer-related information.
- **Home-Grown Status**: Whether the player is home-grown or not.Would you like to focus on any specific aspect for the team analysis? For example, we could look at:
1. Overall team strength based on skill ratings.
2. Age distribution.
3. Strengths and weaknesses in specific positions.
4. Financial aspects like wage distribution and player values.Please let me know how you would like to proceed!
Quote### Overall Team Strength Based on Skill Ratings
Here's how Oxford City fares in various skills, based on the mean ratings of all players in the team:
#### Top 5 Strengths:
1. **Natural Fitness (Nat.1)**: 11.0
2. **Aggression (Agg)**: 10.2
3. **Acceleration (Acc)**: 9.9
4. **Jumping Reach (Jum)**: 9.7
5. **Pace (Pac)**: 9.2#### Bottom 5 Skills:
1. **Eccentricity (Ecc)**: 2.4
2. **One on Ones (1v1)**: 2.4
3. **Throwing (Thr)**: 2.5
4. **Communication (Com)**: 2.6
5. **Kicking (Kic)**: 2.8#### Observations:
- The team excels in physical attributes like natural fitness, acceleration, and pace, which suggests that they would be well-suited for a high-tempo, pressing game.
- The team also has a high mean rating for aggression, indicating a potentially aggressive playstyle.
- However, the team seems to lack in more specialized skills like eccentricity, one-on-ones, and throwing.Would you like to dive deeper into any of these aspects, or would you like to explore something else?
QuoteCreating a tactic in Football Manager 2023 involves various aspects like formation, player roles, team instructions, and set-piece strategies. Based on the Oxford City team's strengths and weaknesses, here's a suggested tactic.
### Formation: 4-2-3-1 High Press
#### Why 4-2-3-1?
- This formation allows the team to make the most of their physical attributes like pace, acceleration, and natural fitness.
- It provides a good balance between defense and attack.
- Utilizes wide players, which is beneficial given the team's strengths in acceleration and pace.#### Player Roles:
- **GK (Goalkeeper)**: Standard Goalkeeper
- **RB & LB (Full-backs)**: Wing-Backs (Support) — to make use of their good acceleration and pace.
- **CBs (Centre-backs)**: Ball-Playing Defenders — to initiate play from the back.
- **CDM (Central Defensive Midfield)**: Deep-Lying Playmaker (Defend) and Ball-Winning Midfielder (Support)
- **CAM (Central Attacking Midfield)**: Advanced Playmaker (Attack) — to make use of their creativity and vision.
- **LM & RM (Left and Right Midfield)**: Wingers (Attack) — to capitalize on their pace and crossing ability.
- **ST (Striker)**: Complete Forward (Support) — a versatile role that can drop deep, lay off passes, or go for goal.#### Team Instructions:
- **Mentality**: Positive
- **In Possession**: Higher Tempo, Play Out of Defense, Use Offside Trap
- **In Transition**: Counter-Press, Counter
- **Out of Possession**: Use Offside Trap, Higher Line of Engagement, Higher Defensive Line#### Set-Piece Strategies:
- **Corners**: Aim for the near post as the team has good aggression and jumping reach.
- **Free Kicks**: Mixed delivery to make use of the team's aggression and heading ability.#### Strengths:
- The tactic makes full use of the team's high natural fitness, aggression, and work rate.
- The high press and counter-press suit the team's overall aggressive nature.
#### Weaknesses:- The high line could be vulnerable to long balls and pacey strikers, so the central defenders should be cautious.
Would you like to know more about how to implement this tactic in Football Manager 2023?
0 -
7 hours ago, toolkit68 said:
Mate - if a specific wording or sentence flips you of be so polite to use quotes and dont just use 'words' or 'statements' - this is the internet! You can link to someone else's rubbish directly, marvellous isnt it?
That is my original statement by the way. If you related to anything else please use the quote function.
No need. You know what I am referring to; don't be coy. But thank you anyway for clarifying that you have no clue and are just giving an opinion.
I also appreciate the lesson on internet etiquette.
But as a fellow developer, seeing the lack of a growth mindset is discouraging. You completely dismissed this feature without trying or giving constructive criticism,
0 -
5 hours ago, toolkit68 said:
Everything in my posts is based on SI's own descriptions in the manual , in dev posts, tooltips ingame, descriptions both in game and so on. The roles of all directors is shown in tutorials, in tooltips and so on. Have you ever read anything in the game actualllay, because your question for a source feels like you have not.
Edit: And if you make an assumption based of my post that i am a developer you are both right and wrong at the same time. I do develop software but I do not work for SI. I post as a normal player who has been playing the different versions of FM for years now.
This doesn't explain your commentary on the last part of your original statement. Seem more speculation.
0 -
1 hour ago, Colonel Tigh said:
Really liking the look of this year’s game. My final wish is to fix the fitness levels during club World Cup /international competitions. I think international football can be completely revamped but that would have been a headline feature so will settle for just fixing the fitness levels
Hopefully, this will be a part of the quality-of-life upgrades. International management being untouched would be a disappointment, considering this is coined as the most complete FM.
0 -
It Just got more valuable with ChatGPT4 using Plus. It now has browser access, so it can use real-time data (which includes forums and sites with tips on FM) instead of what it was trained on up to 2021. It also has a data analysis tool that can break down data and even create charts and graphs...
In short, you can feed it data from your team so it can give you more accurate suggestions.
0 -
57 minutes ago, toolkit68 said:
Source of what? Maybe try two or three words in a row? It is called sentence for a reason
Can you provide sources for your post? Your statement sounds very authoritative and factual rather than being just an opinion. If I didn't see your name, I would have assumed that you were a developer.
0 -
On 25/09/2023 at 11:38, toolkit68 said:
You misunderstand the point:
The offloading of players falls into a certain responsibility area which is usually lead by the Director of Football (some call them Director of Sports, Technical Director and so on but for the FM argument this is what the DoF is for).
The usage of intermediaries is not questioned in general but the implementation is. At best this feature is a sub-button on the DoF UI when you place players on the 'unwanted' list - it is not a major feature as SI makes it out to be. Instead of fixing the bugged offloading this feature is promoted like it is the real solution to a problem - which it isnt.
Source?
0
OpenAI ChatGPT is a gamechanger for FM immersion
in Football Manager General Discussion
Posted
The new feature with web browsing functionality lets it fact check itself by checking multiple reliable sources.