How come nobody is discussing about the AI that's been fixed ?

[Imari]

I think they're just testing the waters with some minor changes every few updates, that's why they haven't listed it in official changelog

It's also possible that they haven't changed anything at all. Racing is a chaotic situation and there's going to be a fair amount of variation in the "quality" of the AI racing even within a single patch.

When new patches drop people are scrutinising every last detail, partly because they want it to be better and partly because we can't trust Polyphony to tell us about any changes that they do make. It's not unlikely that a some people get a few good races in a row directly after the patch and assume that means that there were positive changes made.

To be clear, that's not an unreasonable assumption given that information. Humans are great at detecting patterns even when there's nothing there, that's why we see faces in clouds. But it's a complex enough situation that a few good races probably isn't sufficient to make concrete statements about whether the AI has changed at all, let alone how it has changed. Add in the fact that they could be doing stuff per track, per car or per event and it becomes very, very hard.

The sensible way to approach it is to assume that they haven't changed anything unless someone can gather enough evidence to establish clearly that they have. That's not as exciting as believing that Polyphony are constantly tweaking one of the weakest areas of their game, but it's also more likely to be correct.

 

[Mudda Chod]

This is what I noted in the Custom Race thread, after the August Update. Again, it’s PD being track and car specific.

I’m on PS5. The Tokyo 600 race still have AI taking me out instead of trying to avoid me. However, AI are being more courteous and opening up the corner when, I guess, it senses I’m much faster. This happens at various circuits, but it’s sometimey.
I also have a seven Lap video at Fuji with me not passing many AI from the back. Cars not falling back and actually challenging me for position. Cars are similar horsepower and PP, with some as duplicates with different liveries. I’ll probably post that in the Custom a race thread.

I feel it’s best to test this in Custom Race than PD events. Those can be scripted and too varied. We just don’t know what programming they’ve done for tuning and AI level.

I agree. It appears to be situational. Sometimes the AI are totally courteous, and at other times they are insanely dirty.

Perhaps they have some kind of internal "Aggression Level" that is randomly variable (like weather) at the start of each race.

Before I posted this thread I tested the same race with different cars and on different difficulties and the AI was wonderfully courteous every time, and (like @Nightwing24 mentioned above) some specific movements of the cars at the start of the race were different.

It's a good idea to test it in Custom Races, but the payout is so poor. At least if I do the Tokyo grind I get a decent Cr reward. If/when I notice that the AI are different I'll save the replay and make a video comparison.

It's also possible that they haven't changed anything at all. Racing is a chaotic situation and there's going to be a fair amount of variation in the "quality" of the AI racing even within a single patch.

When new patches drop people are scrutinising every last detail, partly because they want it to be better and partly because we can't trust Polyphony to tell us about any changes that they do make. It's not unlikely that a some people get a few good races in a row directly after the patch and assume that means that there were positive changes made.

To be clear, that's not an unreasonable assumption given that information. Humans are great at detecting patterns even when there's nothing there, that's why we see faces in clouds. But it's a complex enough situation that a few good races probably isn't sufficient to make concrete statements about whether the AI has changed at all, let alone how it has changed. Add in the fact that they could be doing stuff per track, per car or per event and it becomes very, very hard.

The sensible way to approach it is to assume that they haven't changed anything unless someone can gather enough evidence to establish clearly that they have. That's not as exciting as believing that Polyphony are constantly tweaking one of the weakest areas of their game, but it's also more likely to be correct.

True. I posted this immediately after noticing the "change", assuming of course that it was new and repeatable. Now it seems that the "change", if any, is more subtle and elusive, and as you infer, may have already existed before the update. Nevertheless, since others have also observed a considerably different AI behaviour at various times, we are now attempting to "gather enough evidence to establish clearly" the situation with the AI.

 

[havermoutpapi]

yes!!! Thank god and i am glad PD is implementing such necessary changes. I still have hope that they wont give up on GT7 and us

 

[Imari]

True. I posted this immediately after noticing the "change", assuming of course that it was new and repeatable. Now it seems that the "change", if any, is more subtle and elusive, and as you infer, may have already existed before the update. Nevertheless, since others have also observed a considerably different AI behaviour at various times, we are now attempting to "gather enough evidence to establish clearly" the situation with the AI.

It's still good to have mentioned, because now it gets more people looking and gathering data. You're going about it the right way.

We all hope that they're changing the AI, I just know that there's a bit of a tendency of this forum (in general, not yourself) to assume that a change has been made and then look for data that confirms what they think. This has been an issue since GT5 and online updates were a thing, so I just wanted to make clear in the thread that "no change" is also an option in these cases.

I think a lot of the problem is it's hard to have replicable tests. If Polyphony are changing the AI per event we'll probably never nail it down. But if it's a general AI change, then Custom Events like suggested by @05XR8 are probably the go.

Maybe even making them one make and a small field to reduce the variables involved might help. A 2 car race might be the most basic test, as you can put yourself in various positions with respect to the AI and see how it reacts without other cars coming in and spoiling the experiment.

 

[Mudda Chod]

You're going about it the right way.

"no change" is also an option in these cases.

Maybe even making them one make and a small field to reduce the variables involved might help. A 2 car race might be the most basic test, as you can put yourself in various positions with respect to the AI and see how it reacts without other cars coming in and spoiling the experiment.

-Thanks

- Agreed

- Yes, I think that's the kind of reasoning he's suggesting, would make it easier to evaluate the AI in repeatable situations

 

[05XR8]

It's still good to have mentioned, because now it gets more people looking and gathering data. You're going about it the right way.

We all hope that they're changing the AI, I just know that there's a bit of a tendency of this forum (in general, not yourself) to assume that a change has been made and then look for data that confirms what they think. This has been an issue since GT5 and online updates were a thing, so I just wanted to make clear in the thread that "no change" is also an option in these cases.

I think a lot of the problem is it's hard to have replicable tests. If Polyphony are changing the AI per event we'll probably never nail it down. But if it's a general AI change, then Custom Events like suggested by @05XR8 are probably the go.

Maybe even making them one make and a small field to reduce the variables involved might help. A 2 car race might be the most basic test, as you can put yourself in various positions with respect to the AI and see how it reacts without other cars coming in and spoiling the experiment.

-Thanks

- Agreed

- Yes, I think that's the kind of reasoning he's suggesting, would make it easier to evaluate the AI in repeatable situations

For sure. I did some races with just the Gran Turismo standard six cars in GT Sport, but the cars were not equal. Now, with users being able to assign computer cars, it makes testing easier.
Just a side note again, with PD changing the suspension settings to mimic Gr.3 & Gr.4 default settings, it also helps to get a base level or what the AI are “used to”. Meaning, AI handling should be consistent, even with AI using all aids. Having none of the weird angles we see with stock suspensions. It reads a bit silly, but it’s a variable I’ve taken out, by setting my suspensions just like those cars Gr.3 & Gr.4 cars. The default frequency are softer than the race cars, but it’s funny PD chose these particular settings.

I understand everyone go about setting up their own way. This is just what I do.

 

[fastone371]

I agree. It appears to be situational. Sometimes the AI are totally courteous, and at other times they are insanely dirty.

Perhaps they have some kind of internal "Aggression Level" that is randomly variable (like weather) at the start of each race.

Before I posted this thread I tested the same race with different cars and on different difficulties and the AI was wonderfully courteous every time, and (like @Nightwing24 mentioned above) some specific movements of the cars at the start of the race were different.

It's a good idea to test it in Custom Races, but the payout is so poor. At least if I do the Tokyo grind I get a decent Cr reward. If/when I notice that the AI are different I'll save the replay and make a video comparison.



True. I posted this immediately after noticing the "change", assuming of course that it was new and repeatable. Now it seems that the "change", if any, is more subtle and elusive, and as you infer, may have already existed before the update. Nevertheless, since others have also observed a considerably different AI behaviour at various times, we are now attempting to "gather enough evidence to establish clearly" the situation with the AI.

Most of my experience with AI is at the Tokyo grind. It seems to me that maybe PD added a couple more scripts for AI cars. I still see them suddenly change lanes in front of you when you go to lap them.

 

[daan]

Am I supposed to be psychic to know that?

No, but you're bringing up the supposed slow AI in a race that has some of the fastest AI in the game, and if you'd played the game, you would know that. GT7 has many faults but it doesn't need someone who has never played it kicking it every opportunity they get. You'll even kick it when there isn't an opportunity just like you've done here.

I am very curious what is your definition of glitch because for me it seems that anyone who can tune properly car (with what PD gave us) is using a glitch 🤣

So the Aston DP100 is also a glitched car?? I regularly lap at least the top 14 cars and sometimes the whole field using it. I have not seen PD do anything to penalize the DP100 yet.

I thought he was the one doing the rotary Cappucino thing? That hardly seems like a glitch. It's a car with an engine swap and a tune.

Probably not glitched, but there are definitely cars that break the PP system.

 

[Imari]

Probably not glitched, but there are definitely cars that break the PP system.

Sure, if you define "break" in a very specific way. But much like any system of rules regarding performance in motorsport, half the fun is figuring out how to get a car that goes really fast while still being technically within the letter of the law.

It's "broken" in the sense of being really good, like Meta Knight in Smash Brothers Brawl. It's not "broken" in the sense that the system isn't working as intended. Should it be removed or patched? Maybe, but it is pretty fun given that it's mostly applicable to single player.

I'd say that it's not the player's fault if Polyphony can't figure out how to create a system that accurately represents the performance of a given car, but to be fair to them it's an impossible problem. The idea that you can have one number to try and quantify all aspects of performance of a car in a realistic simulation is ridiculous. If it was that easy then car design would have been solved decades ago. On any given track it's inevitable that there's going to be a huge range of lap times for different cars at a given PP simply because of how the system seems to work.

The rotary Cappuccino is the side effect of this. It's working exactly as intended, it just happens to be a really good combination.

Whatever you want to call it, you seem to keep trying to imply that there's something wrong with using a car that is unquestionably fast for it's PP value. And that's just not true, it's arguably the whole intent of the tuning system which is in turn a fair analogue for designing a race car in real life. Finding a car, a set of parts and a tune that are a great match for a given track is what it's about.

 

[Fuz_Belvedere]

As someone that races solely against AI, nothing is fixed... although they have increased the pacing of the nearest cars (even with boost off) and turned up the player catch-up ability even on (soon to be) lap down cars.


I can not wait for the Sophy AI upgrade (hopefully with a % adjustment)

 

[RallycrossGT]

I don't know, man... I don't really race Tokyo a lot but I had a few goes yesterday and I still think...

• AI brakes in weird places
• Will still pit maneuver you if too close
• Will still plow through you if they are faster on the straight (especially the damn GT-R)
• Contact physics don't work for them

 

[GTboyz]

the AI in Tokyo might be a BIT better but still kinda crap.

At Le Mans though, brake checks everywhere.

 

[Imari]

I can not wait for the Sophy AI upgrade (hopefully with a % adjustment)

I wouldn't get your hopes up that you'll be able to race a field of Sophys in GT7 in career or custom races. I'd give it about 50:50 that the tech makes it into the game at all, and if it does there's a pretty good chance that it's a cut down version that only works in specific 1v1 challenges or something weird. It's highly unlikely to just drop in and make all the AI better.

Perhaps with GT8, but even then the whole machine learning thing seems like marketing spin. As far as having AI that can race well against non-aliens, that exists with current technology and has done for some time. Polyphony just choose to put their resources towards other things.