enigmatic

The thing about targeting players by media description is that since it's reputation based you'll never be able to afford them all, will spend a lot of money on some CA 130 teenagers that are playing really well but have a CA of 138 and will still miss some sensations. The NxGn lists are very hit and miss too. For realism purposes that's great (and you'll also mop up a lot of PA155 players that can forge a great side). But if you actually want to focus on the Messis, use an editor or third party "scout" tool and prepare to be surprised I mean, here's a player who's much better than most players his age, at Man Utd and has already played in the Europa League. He also has 200PA and a personality that will probably let him get there, so he's definitely the Messi of his generation. His description is only "attacking midfielder" though (Yes, I peeked. First save I ran when the game was out, mainly for test purposes and with an obviously stacked club, so it wasn't terribly serious. Might go back and play it at some point though, simply because I doubt I've ever had a youth prospect this good in normal play)

The unrealistic bit is getting Villa as a senior affiliate... (He's not interested so won't go to you anyway)

Well if it's so easy, you could always do it yourself! In practice of course, it isn't that easy at all. There's an enormous amount of processing power required to run an LLM, and an enormous number of man hours to debug it. OpenAI is spending much, much more than SI's development budget, and SI have to maintain an actual game too! I mean, building a context sensitive help tool which uses something like Dialogflow to link human questions to articles about how to play is pretty easy, but I'm not sure making FM an online only game with a subscription based on how much you play just to have an FM version of Clippy would be that popular. (One fansites could implement if they wanted to pay for the API calls though...) And in case you haven't looked at the examples here, ChatGPT is tediously repetitive (unless you choose to do outlandish things like ask for a match report in the style of the Bible), it's just tediously repetitive and erratic and difficult to debug. By the time you've trained it (and turned down the temperature setting enough) to be accurate enough to be useful, it's probably not much different to the current limited interactions. tbf to ChatGPT, this tendency to be overly wordy is mostly because human testers are more impressed with verbosity and appending lots of caveats (including redundant ones) than concise responses. I'm not sure how much of this is because people like waffle in general, how much is because maintaining coherence whilst being longwinded is seen as an accomplishment for a computer, and how much is because the waffle and the caveats cover for pretty dubious answers.

It literally didn't. The word "defend" appears in it twice, once in reference to set pieces and once in reference to them forming a compact block, which it explains in more detail in the next section. It's a mid block, to block passing lanes in the centre of the pitch, not a low block deep in their own half. They had Rio Ferdinand and Patrice Evra ffs, they didn't need to crowd the penalty area to control the space in behind! Again, if you are a bot who has never watched Man Utd and doesn't know what a mid block is, it's understandable you link the word "defend" to "defend deep in own half". But they are not the same thing at all.

It doesn't have statistics or data on the game. It has lots and lots of text, and the ability to write more text that looks similar I'll say one thing for ChatGPT though, it does try to learn from people's text rather than just saying "now you're just arguing for the sake of arguing" when people who know much, much better than they do how LLMs work try to explain how it works and why it comes up with nonsense like Depth: Shallow (hint: it's because it maps very well to Width: Narrow in a word model, just not at all in a "how does football manager work" model...) You seem to be very stuck pretending that you didn't quote ChatGPT saying The team would be instructed to defend deep in their own half, looking to soak up pressure , y'know, the two phrases I quoted back when pointing out how that advice was wrong. Maybe you're an LLM and you're just designed to emit words whether you understand the conversation or not? The whole idea of a help tool is that you're supposed to be using to get help, not trying to help it A beginner doesn't mean an idiot, but it does mean someone that's going to spot subtle problems with the suggestions and be confused by the stuff that looks like tactical instructions but is actually just words statistically similar to tactical instructions... Glad they've finally tweaked it so it stops suggesting standard formations when you have a man sent off though; that's actually an interesting discovery. Last time I tried that was yesterday, so it's progress Still, I just told it about having two players sent off and.... I am winning 1-0 but my right back and right wing are sent off. What formations could I play? When you're winning 1-0 but your right back and right wing are sent off, it can be a difficult situation to manage. Here are a few formation options you could consider in this scenario: 5-3-2: This formation would involve adding an extra defender and dropping one of your forwards. This would give you more defensive cover on the right side and make it harder for the opposition to break through, while also allowing you to maintain your lead. 4-4-2: This formation would involve dropping one of your forwards and pushing one of your midfielders back to play as a defensive midfielder. This would give you more defensive cover on the right side and make it harder for the opposition to break through, while also allowing you to maintain your lead. 3-5-2: This formation would involve adding an extra defender and dropping one of your forwards. This would give you more defensive cover on the right side and make it harder for the opposition to break through, while also allowing you to maintain your lead. 4-3-3 (Diamond Midfield): This formation would involve dropping one of your forwards and pushing one of your midfielders back to play as a defensive midfielder, and also dropping another forward. This would give you more defensive cover on the right side and make it harder for the opposition to break through, while also allowing you to maintain your lead. 3-5-1-1: This formation would involve dropping one of your forwards and pushing one of your midfielders back to play as a defensive midfielder, and also dropping another forward. This would give you more defensive cover on the right side and make it harder for the opposition to break through, while also allowing you to maintain your lead. It's worth noting that the best formation for you would depend on your team's strengths and weaknesses, and the opposition you're playing against. It's also important to note that you'll also have to adjust the other players roles and duties to support the new formation, and use the team instructions and player instructions offered by Football Manager to make the most of your new formation and players. The key is to maintain a balance between defense and offense, and to be able to keep the ball and create opportunities while being able to defend properly. You should also consider the time left in the game, and make sure that the formation you choose allows you to protect your lead and avoid conceding goals. This is what you expect from a model that associates "dropping one of your forwards" and "more defensive cover on the right side" with "right wing sent off", and has a good model of how to format a list of formations. It even manages to associate "adding an extra defender" with 5-3-2 and 3-5-2, which would be impressive except for the very important point I haven't told it how many defenders I actually have... It's not what you'd expect from a model that has a grasp of what formations and sendings off actually are...

The bit that made me laugh is when a manager threw his toys out the pram and recalled a Backup player that was getting regular football in a variety of positions because only some of them were in his preferred position, whilst nobody bothered about all the other loans I had who were alienated because they weren't getting anywhere near as much football as expected...

Apart from the bit where you're insisting that ChatGPT isn't automatically generating responses but is doing "research" and pulling "data" and I'm talking about it being a word model with automatically generates responses based on text content I'm saying "literally the exact same thing" yeah, sure... (incidentally, confidently insisting that two completely different and contradictory things are the same is one of ChatGPT's major weaknesses....) It's a sentence generator, which sees words and responds with more words statistically related to those words in novel sentences. It's a lot better at this than a Markov chain, but the extent of its "knowledge" is pretty much limited to (sometimes) being able to apply basic logic to specify the correct number of players if you ask for a back four. Still, if ask it what formation to play after you've had a man sent off, it'll write some stuff associated with "man sent off" about defending more and then give you some words it knows which are different formations which all have 11 players Sure, if you tell it you have some tactical problems it will - with no understanding of your tactic and how the different positions interact - write some sentences vaguely associated with tactical problems. Sometimes it might hit on one which is actually useful. But so can writing all the possible tactical changes and rolling a dice. Advantage of the latter is that it won't tell you to use tactical options that don't exist like setting Depth: Shallow! You've given us quite a few examples of specific questions, half of which had specifically wrong options I asked some specific questions about some silly formations myself and it wrote me a nice essay about changing roles in a "standard 4-4-3 formation" (it was actually 5-5-3 tbh....) and in a formation with no central defence or midfield it suggested the FB(S) role was too attacking... A good example of a blog which gives much more detail on how Man Utd played and would be far more useful than ChatGPT as a basis for trying to create your own version of Man Utd's 2007-9 era formation, yes. Things the blog doesn't suggest: that they played with a deep defensive line to soak up pressure. Because nobody who watched Man Utd play that way would suggest they were noted for setting up with a deep defensive line to soak up pressure (especially not how FM interprets deep defensive lines) I said the tactic was usable (if you ignore the bits of it which are wrong like a back four with wingbacks in the wingback position). But having a DLP as a primary playmaker and the AP in a wide role which wasn't even called "Advanced Playmaker" until recently is a pretty bad suggestion for understanding how to build a team around an Advanced Playmaker.

Respectfully, I know a lot more about how Large Language Models like ChatGPT work than you do. It doesn't have "data" on how your tactic behaves or do "research", it's a big statistical model based on a lot of sentences previously written by humans, some of which were on the subject of Football Manager and football tactics in general, and so it can write similar sentences involving similar words which are statistically associated with your question, which may or may not have any correspondence to the actual thing that's making you lose games and might be complete nonsense, like telling you that your narrow width is complemented by shallow depth. (and FWIW it's much better at doing simple one-hit syntactic operations with code than high level reasoning about it) .The team would be instructed to defend deep in their own half, looking to soak up pressure2 Watched Sir Alex Ferguson's Man Utd literally hundreds of times, and I think he picked this as their starting defensive setup twice (home leg and away leg semi finals against Barcelona in 2007/8). For obvious reasons, he didn't combine this with attacking fullbacks and a front two. But ChatGPT doesn't get to watch Man Utd or play football manager, it just "knows" that Alex Ferguson is correlated with references to counter attack, and counter attack often involves instructions to "defend deep in their own half" and "soak up pressure", puts 2 and 2 together and makes 5 Lots of blogs by people that have watched Man Utd and played Football Manager do a better job. Ironically Siri can probably find you some... It's actually pretty bad (supposed to be a back four but also supposed to have wing backs in the wingback position, there's a weird mix of tactical instructions, PPMs, and other words which are sort of similar to tactical instructions and PPMs, and sticking your Advanced Playmaker as part of a front 3 with another playmaker behind is a quirky way of suggesting to use it). Not completely unusable like the Fergie formation, but I'm not sure it's any better than what a typical novice FMer might pick for themselves...

It is auto generated. It builds a big statistical model of all the stuff in its database so it knows that stuff like "the game Football Manager" is associated with specific phrases like "Play out of defence" which is associated with "Ball Playing Defender which is associated with "Pep Guardiola", and it has a model of how humans phrase stuff which includes enough FM-related content to know how to generate something that looks like a list of positions and instructions What it doesn't know is how to actually play the game, the difference between PPMs and instructions or even that there are supposed to be 11 players on the pitch! It's complete nonsense though. Man Utd have literally never combined two strikers, a very deep defence and a high press, and that tactic would perform dreadfully to the extent it's actually possible to implement. It doesn't pull data. It makes up sentences with words which are correlated with one another Which means that as "counter attack" gets mentioned a fair bit in connection with "Sir Alex Ferguson", and "counter attack" also gets mentioned a lot in connection with "deep defence", it ends up giving you a suggested Man Utd tactical approach which is completely incoherent (despite there being dozens of Ferguson's Man Utd tactical recreations online) You asked it a specific question. It gave you an answer which involved a lot of words which look very much like the football advice that people give except for being specifically wrong on most points. If you struggled with this tactic, it wouldn't tell you it was completely incoherent like people on the tactics forum would, it would suggest some largely random changes.

Sure, if you tell it it's wrong it'll say "I am sorry" and make up some more nonsense that has similar words and structure to people that give tactical advice use. Not really sure that new players are the people who want to be using a tool they have to correct because its advice is impossible to follow though , and lots of the stuff is subtly wrong (you *can* set up to defend deep in your own half, but you certainly shouldn't do that if you're trying to emulate the tactics of Sir Alex Ferguson, so if you're looking for advice, you'll end up worse off than if you tried to figure it out yourself Can't see why any new player would want to use a ******** generator rather than searching for advice written by humans that have actually played the game. Useful for entertainment if you want to generate random stories about team that only loosely match the events of the season, not useful for tactical advice except for amusement purposes.

It's not instant data if it's telling you to select tactical options and roles that don't actually exist though, like a Balanced Team Shape, a Shallow Depth, a Box to Box midfielder on the right wing and Make Runs Behind Defense as a tactical instruction.... In fact, I think fully half the things it's suggested you do for the 4-4-2 aren't actually possible...

tbf this one has improved quite a bit in recent versions: you get a promotion-sized transfer budget and reputation boost as soon as you secure promotion which is better than waiting until season end

The ChatGPT API is temporarily free for research purposes. They're not giving uncapped API access away to the developers of a game that sells millions of copies and primarily trains their model on arcane football terms of limited wider interest. And it'd be unusable without a stable internet connection, which is a pretty major change on how FM currently works... But even if it was free and easy it integrate, the feedback it provides isn't "meaningful" because it has no idea of how to play the game and no ability to impact the game, and it'd also be essentially impossible to debug. It provides stuff which looks like meaningful feedback because its paragraphs based on words you use and words which commonly appear in football blogs, but is actually complete nonsense, like suggesting you don't mark the keeper to avoid giving away free kicks near the edge of your goal, players should lurk at the far post at corners because they're good at corner taking or long shots, or suggesting to spend 40% of your budget paying your star player 45000 and 27% of your budget paying the next three players 30000. Likewise, it can write all the autogenerated summaries and questions in a press conference you want, but it can't make anything you say actually have an impact on players or club management. Why would SI spend their money on something which is just going to multiply complaints about bugs by 10...

The "badges" are just game-y difficulty levels to set your reputation and stats and give you in-game options to buff them Of course IRL you'd have a qualification even if taking on a lower league team, you wouldn't be allowed to manage at senior levels without UEFA licences and even if they were in financial difficulty the course fees wouldn't put clubs off (even the UEFA Pro is around €10k) or stop managers acquiring them themselves if the club insisted it wasn't going to pay You can start off as manager of Benfica or Porto with a Continental Pro license and Sunday League playing experience if you want. That's actually lower playing experience than Jose, skips his years as assistant manager and you can even start in the Primeira Liga manager without the badges which wouldn't even be allowed IRL

tbh I tend to do that with youngsters. But the obvious answer to that question is "because to get them actually good on the opposite side of the pitch takes a long time and you have to actually play them there even though they're not naturally suited to it" either way, as far as I can see the CA weighting algorithm doesn't affect the RCA at all if you edit a DL to be equally adept at playing DR for instance. The official answer your colleagues have always given is that the ability to play multiple positions itself isn't supposed to add anything to CA by itself, it's just a side effect of CA being the sum of attributes weighted by the *highest* attribute weight associated with any of the positions they play, which means once your fullback with good finishing learns to play winger, his finishing takes up a lot more CA than when he was only a fullback and his finishing didn't matter that much...

Most of what reputation counts for is AI tactical choices. And generally the underdog human player has a massive advantage over the AI underdog because its tactical choices aren't constrained to picking realistically defensive formations which lose a realistic amount of the time...

Maybe your tactics weren't as balanced and your team not as strong as you thought! It's not like Liverpool and Real Madrid aren't frequently beaten by human managers in game. Don't think I've ever managed to lose 9-0 either, and I've played against a lot of teams that are better than me including elite ones, and occasionally even got a bit silly trying to chase the game against them. If anything I'd say FM typically has the opposite problem: high reputation AI sides underperform on average against human sides because much of their success against weaker AI-managed sides comes because the weaker sides are set up to keep the score down rather than exploit their weaknesses.

If you concede 40 shots with 11 men on the pitch against a team that isn't a couple of divisions better than you, the problem isn't that there's some magical easter egg the developers put in to make you personally outclassed by Real Madrid (but none of the many, many FM players that beat Real Madrid) the problem is that you screwed up your tactics. Period.

If you want to try to find the kryptonite tactic for a particular tactic, you could always set the opposition up to play that yourself rather than expecting SI to write illogical logic to do it...

This. Huge amount of work involved in developing a system that figures out what the "best" human tactics are, embeds them in the game and assigns them to managers at appropriate times, and if you get it wrong league tables start to look silly. Why would you invest all that time and money for a checkbox option for a literal handful of people that actually want to play against random plug and play exploit tactics rather than the actual range of tactical decisions a manager would be likely to make in a game. Especially when it goes against the whole ethos of FM (simulating football, tactics being something you adjust to the match situation not "plug and play", exploits being considered areas for improvement in future match engines rather than a desired feature of the game)

Have you looked at the real life figures for managers dumped (or left for new job, which is included in your figures)? They tend to be very similar, sometimes actually higher It's not surprising that SI isn't responding if you haven't actually provided any evidence there's a problem. Think 16 managers being sacked from Serie B in game is "some sort of joke"? Looks like 14 got sacked during the course of the 2021-22 season proper and another 10 replaced at the end of the season last year, with Cosenza alone having four managers (who judging by the timings weren't intended to be caretakers!)

No. (I'd explain, but your own post already does that )

The AI randomly using player tactics seems to sort of miss the point of FM being a simulation of the world of football though. I'm sure AI Pep could win even more consistently using ChineseForumChickenWing523CornerExploit but it sort of misses the point that this isn't how he actually plays...

I do wonder whether any of the people complaining that sometimes they take more shots than the opposition to score have ever watched a game of football before

Don't think there's anything anomalous about a senior FIFA exec and a 65 year old former big name manager whose last role was running an academy 8 years ago not being available to hire as coaches or applying for managerial jobs, and Redknapp said he'd retired from management in 2017 and is now 75. The bigger anomaly might be people like Taribo West still being in there. Some people probably stay in there simply because nobody's paying any attention to them, and he might be one of them. Apparently he now runs a church! McBride has a staff role with the US men's national team though