Is Sophy 2.0 no longer competitive?

[Artremis]

Did the old AI opponents in Arcade Races have similar slowdowns? I wonder if this is something being applied to Sophy's vehicles by the game mode, and not Sophy itself doing this. I guess an easy test would be to see if the non-Sophy tracks in Quick Races have the same AI slowdowns.

They do. An easy way to test it in a custom race is to set a 2-3 lap race in a car that's somewhat slower than the rest of the grid. You will notice that on lap 1 you will simply get passed for being too slow, then magically in lap 2 all of the other cars slow down and you can do overtakes again.

But since it's possible custom races use a different setting, I tested it on a quick race with Gr.2 cars in Deep Forest. And sure enough:

 

[GTboyz]

What if it has something to do with PD doing 16 cars instead of less ?

 

[ShiftingGears]

People are just using suped-up cars against the AI and then wondering, "Why is the AI so slow?"

You seem to be mixing "people are using souped up cars" with "people are not actively handicapping themselves" here.

Ultimately, player and AI cars being equal, Sophy is genuinely not very challenging by a lot of people's standards.

You can take any group 3 car, put matching tyres (racing hards) on, drop to last, then re-pass the entire field and win the race. The AI is handicapped on power while it is ahead of you.




The 911 RSR has 500 horsepower which is not a lot compared to the rest of the Group 3 cars, and it still closes on every single opponent on the straights, and not just because of the corner exits.

On the run down to Eau Rouge I gain two tenths on the McLaren F1 GTR, which has 100HP more and 200kg less than the car I am using.

As soon as I pass the leading Jaguar, all of a sudden it is faster than me on the straights. The Jaguar's lap time on the final lap is five seconds faster than their time on lap 2.

Conveniently leaving out the part of them being in an overpowered car.

Every car is overpowered against Sophy - while you are behind them. Thats how they've dumbed it down for this iteration.

 

[Yard_Sale]

You seem to be mixing "people are using souped up cars" with "people are not actively handicapping themselves" here.

Ultimately, player and AI cars being equal, Sophy is genuinely not very challenging by a lot of people's standards.

You can take any group 3 car, put matching tyres (racing hards) on, drop to last, then re-pass the entire field and win the race. The AI is handicapped on power while it is ahead of you.




The 911 RSR has 500 horsepower which is not a lot compared to the rest of the Group 3 cars, and it still closes on every single opponent on the straights, and not just because of the corner exits.

On the run down to Eau Rouge I gain two tenths on the McLaren F1 GTR, which has 100HP more and 200kg less than the car I am using.

As soon as I pass the leading Jaguar, all of a sudden it is faster than me on the straights. The Jaguar's lap time on the final lap is five seconds faster than their time on lap 2.


Every car is overpowered against Sophy - while you are behind them. Thats how they've dumbed it down for this iteration.

What difficulty level were you playing on?

 

[ShiftingGears]

What difficulty level were you playing on?

The hardest difficulty.

 

[Yard_Sale]

The hardest difficulty.

Copy

 

[Bojador]

It’s a catch-up mechanic. Alex Guilon has a video on YT and it would seem that if you slow down the AI slows down as well and takes a good 15 seconds extra to finish a race around Tsukuba on gr1. If you drive normally they won’t slow down as much.

This is what infuriates me about PD. They have access to this new ground breaking AI and they choose to make the hardest difficulty easy with this kind of mechanics.

Has anyone tried the easier difficulty? Maybe they messed up again and it’s actually the hardest. (I know I’m on denial).

Can you imagine if they took one of the best races in the game (1h Lake Magiore or Suzuka/Daytona GT3 cup) and just use Sophy? That would actually be a good showing of Sophys capabilities.

 

[rlx]

Tested Sophy 2.0 again on hardest difficulty at Le Mans, this time in an engine-swapped Golf GTI.

In the first race, I took P1 immediately, in the second race, I stayed in P20. In the replays, I looked at a Mazda Demio that was present in both races. Listed are top speeds at four points:
before the first chicane / before the second chicane / before Mulsanne Corner / before Indianapolis Corner

Quick Race, me in P1
Lap 1: 183 / 175 / 170 / 178
Lap 2: 181 / 176 / 170 / 178
Lap 3: 181 / 172 / 169 / 177

Quick Race, me in P20
Lap 1: 159 / 154 / 151 / 157
Lap 2: 163 / 154 / 151 / 157
Lap 3: 160 / 155 / 151 / 158

Then, in the pre-race screen (where you see shots of the cars racing on the track), I recorded telemetry data of the vanilla AI driving the same Mazda Demio (stock, Comfort Mediums, PP 322.99) for 3 laps. Finally, I did 3 laps on my own.

AI, pre-race screen
Lap 1: 172 / 168 / 164 / 175
Lap 2: 171 / 168 / 164 / 176
Lap 3: 173 / 168 / 164 / 176

Me
Lap 1: 180 / 173 / 168 / 173
Lap 2: 179 / 174 / 168 / 173
Lap 3: 180 / 174 / 168 / 174

If we take the pre-race screen AI as a baseline, the issue with Sophy is obvious (and it's not a bug where performance would drop over time): it drives reasonably well, and maybe even a little faster than humanly possible, when you're ahead (you won't notice this though, other than in the rear view mirror), and performs astonishingly bad when you're behind.

So the best racing AI ever developed has been neutered to match the behavior that has plagued the vanilla GT7 AI ever since ;-(

 

[Bojador]

Tested Sophy 2.0 again on hardest difficulty at Le Mans, this time in an engine-swapped Golf GTI.

In the first race, I stayed in P1, in the second race, I stayed in P20. In the replays, I looked at a Mazda Demio that was present in both races. Listed are top speeds at four points:
before the first chicane / before the second chicane / before Mulsanne Corner / before Indianapolis Corner

Quick Race, me in P1
Lap 1: 183 / 175 / 170 / 178
Lap 2: 181 / 176 / 170 / 178
Lap 3: 181 / 172 / 169 / 177

Quick Race, me in P20
Lap 1: 159 / 154 / 151 / 157
Lap 2: 163 / 154 / 151 / 157
Lap 3: 160 / 155 / 151 / 158

Then, in the pre-race screen (where you see shots of the cars racing on the track), I recorded telemetry data of the vanilla AI driving the same Mazda Demio (stock, Comfort Mediums, PP 322.99) for 3 laps. Finally, I did 3 laps myself.

AI, pre-race screen
Lap 1: 172 / 168 / 164 / 175
Lap 2: 171 / 168 / 164 / 176
Lap 3: 173 / 168 / 164 / 176

Me
Lap 1: 180 / 173 / 168 / 173
Lap 2: 179 / 174 / 168 / 173
Lap 3: 180 / 174 / 168 / 174

If we take the pre-race screen AI as a baseline, the issue with Sophy is obvious (and it's not a bug where performance would drop over time): it drives reasonably well, and maybe even a little faster than humanly possible, when you're ahead (you won't notice this though, other than in the rear view mirror), and performs astonishingly bad when you're behind.

So the best racing AI ever developed has been neutered to match the behavior that has plagued the vanilla GT7 AI ever since ;-(

We can close this thread now. PD can’t help themselves or they think this is what we want. And maybe this is what the majority of players want.

Thanks for all the tests.

 

[Sir Crashalot]

It's hard to really judge this anyway just alone because of all the different cars used.
If they really wanna show what Sophy is capable of a short 1vs1 with the same car would be the way to go.

BTW: I thought I try the Go Kart because I thought that Sophy would maybe use Karts as well. Turns out Sophy can't drive a Kart😄

 

[Bojador]

It's hard to really judge this anyway just alone because of all the different cars used.
If they really wanna show what Sophy is capable of a short 1vs1 with the same car would be the way to go.

BTW: I thought I try the Go Kart because I thought that Sophy would maybe use Karts as well. Turns out Sophy can't drive a Kart😄

Tried that too. I was hoping that would be a single make race.

 

[Josephs]

For me, Sophy remains a revelation to see lifelike behaviours exhibited by the AI. It's nice to think that regardless of Sophy 2.0 impressions, that future game updates might include notes for improving Sophy (ie: GT7 1.41 + Sophy 2.01) and beyond.

 

[rlx]

It's hard to really judge this anyway just alone because of all the different cars used.
If they really wanna show what Sophy is capable of a short 1vs1 with the same car would be the way to go.

I don't think it's hard to judge, and from what I've seen (and others are reporting), it affects all cars.

PD have shown what Sophy is capable of with the public demo in February, when it was ultra-competitive.

I would understand if, in the process of scaling up from 4 to 19 agents in parallel, they would have had to use a smaller model that is faster to query, but slightly less skilled than Sophy 1.0. But that's not what has happened -- or if it has, it's almost unnoticeable because of the real problem with Sophy 2.0, which is the same old rubber-banding.

I really hope that PD/Sony realize that there are players who want to race against the actual world-class AI opponent that they have developed, and that Sophy 2.0 doesn't deliver on that promise.

 

[Bojador]

I don't think it's hard to judge, and from what I've seen (and others are reporting), it affects all cars.

PD have shown what Sophy is capable of with the public demo in February, when it was ultra-competitive.

I would understand if, in the process of scaling up from 4 to 19 agents in parallel, they would have had to use a smaller model that is faster to query, but slightly less skilled than Sophy 1.0. But that's not what has happened -- or if it has, it's almost unnoticeable because of the real problem with Sophy 2.0, which is the same old rubber-banding.

I really hope that PD/Sony realize that there are players who want to race against the actual world-class AI opponent that they have developed, and that Sophy 2.0 doesn't deliver on that promise.

You would think that’s what the different difficulty settings are for. I can’t even phantom what the easier difficulty is like.

 

[rlx]

You would think that’s what the different difficulty settings are for. I can’t even phantom what the easier difficulty is like.

I tried it. Sadly, they didn't repeat the endurance mission bug: easy is in fact easier than hard :-(

 

[Sir Crashalot]

I don't think it's hard to judge, and from what I've seen (and others are reporting), it affects all cars.

Sorry, I meant it's hard to judge how good Sophy actually is. That would be easier to judge in 1vs1 or at least one make races.

 

[br1x]

Sorry, I meant it's hard to judge how good Sophy actually is. That would be easier to judge in 1vs1 or at least one make races.

I don't think it's very hard at all, it is very obviously handicapped in some form, probably simply in power.
In Gr.2 with the CLK LM and a field with a decent amount of other CLK LM, you just breeze past them on the straights.
In a Gr.3 race with even the Subaru BRZ GT500 you just breeze past them on the straights.
The behavior in defending and trying to re-overtake on the straights is good, they seem to have a lot less power.

 

[newmedia_dev]

I think they have also reworked the old AI as well because on easy they are now comically slow. Well more comically slow than it was.

 

[benjoi84]

I'm no expert, but I would imagine "you have to let your players win, quite a lot of the time" is 'Triple A Videogame Design 101' mantra.

That's why crushingly hard games are often released by indie developers.

I don't understand why we just can't have more difficulty options, though.

 

[wdreamsmaycome]

nobody's saying that Sophy should be impossible to beat, but there should be a way to set it up so that at least it does not artificially slow down and/or rubber band. If they want to tune the difficulty they should just tune its error rate, it's a lot more fun to take advantage if the other car makes a mistake on corner entry or exit as opposed to driving by unopposed on the straight...

I hope once they finally add Sophy to custom races it will be possible to tune it a bit better to make it as hard and/or consistent as one needs to make it an enjoyable opponent and better training for racing online (where nobody's going to lift on a straight to let you go by). If it's a CPU issue I'd rather be able to set say racing only against 4-5 "good" opponents and make that trade off to get the full performance out of it.

I was going to finally upgrade my PS4 to PS5 to get access to Sophy, but if this is the direction PD wants to go with it I guess I won't bother, as I've had enough of "chase the artificially slowing rabbit for 3 laps" type races. It's a pity though, because there doesn't seem to be any game on any platform with comparable AI to race against.

 

[JTB10000]

Beyond disappointing they've applied the same rubberbanding nonsense to the Sophy AI as the regular AI with no option to turn it off. I don't understand this company and so many of its decisions.

 

[tofuhero]

Tested Sophy 2.0 again on hardest difficulty at Le Mans, this time in an engine-swapped Golf GTI.

In the first race, I took P1 immediately, in the second race, I stayed in P20. In the replays, I looked at a Mazda Demio that was present in both races. Listed are top speeds at four points:
before the first chicane / before the second chicane / before Mulsanne Corner / before Indianapolis Corner

Quick Race, me in P1
Lap 1: 183 / 175 / 170 / 178
Lap 2: 181 / 176 / 170 / 178
Lap 3: 181 / 172 / 169 / 177

Quick Race, me in P20
Lap 1: 159 / 154 / 151 / 157
Lap 2: 163 / 154 / 151 / 157
Lap 3: 160 / 155 / 151 / 158

Then, in the pre-race screen (where you see shots of the cars racing on the track), I recorded telemetry data of the vanilla AI driving the same Mazda Demio (stock, Comfort Mediums, PP 322.99) for 3 laps. Finally, I did 3 laps on my own.

AI, pre-race screen
Lap 1: 172 / 168 / 164 / 175
Lap 2: 171 / 168 / 164 / 176
Lap 3: 173 / 168 / 164 / 176

Me
Lap 1: 180 / 173 / 168 / 173
Lap 2: 179 / 174 / 168 / 173
Lap 3: 180 / 174 / 168 / 174

If we take the pre-race screen AI as a baseline, the issue with Sophy is obvious (and it's not a bug where performance would drop over time): it drives reasonably well, and maybe even a little faster than humanly possible, when you're ahead (you won't notice this though, other than in the rear view mirror), and performs astonishingly bad when you're behind.

So the best racing AI ever developed has been neutered to match the behavior that has plagued the vanilla GT7 AI ever since ;-(

Do you know if Sophy actually drives at less than full throttle in subsequent laps, or is the HP of their cars being dynamically limited by the game? I suspect it's the latter, if this same behavior is being applied to non-Sophy AI races.

 

[dabz343]

That would be easier to judge in 1vs1 or at least one make races.

Try the R8 Gordini 1966 at Redbull.

 

[Bojador]

Do you know if Sophy actually drives at less than full throttle in subsequent laps, or is the HP of their cars being dynamically limited by the game? I suspect it's the latter, if this same behavior is being applied to non-Sophy AI races.

As far as I tested they apply full throttle, contrary to what happens in custom races where the AI does not apply 100% throttle.

So they are being limited by some other way. If the HP was limited they would reach the same top speed every lap so I assume is the good old boost (or in this case the opposite of boost, handicap?).

 

[Dena RACING]

Tested Sophy 2.0 again on hardest difficulty at Le Mans, this time in an engine-swapped Golf GTI.

In the first race, I took P1 immediately, in the second race, I stayed in P20. In the replays, I looked at a Mazda Demio that was present in both races. Listed are top speeds at four points:
before the first chicane / before the second chicane / before Mulsanne Corner / before Indianapolis Corner

Quick Race, me in P1
Lap 1: 183 / 175 / 170 / 178
Lap 2: 181 / 176 / 170 / 178
Lap 3: 181 / 172 / 169 / 177

Quick Race, me in P20
Lap 1: 159 / 154 / 151 / 157
Lap 2: 163 / 154 / 151 / 157
Lap 3: 160 / 155 / 151 / 158

Then, in the pre-race screen (where you see shots of the cars racing on the track), I recorded telemetry data of the vanilla AI driving the same Mazda Demio (stock, Comfort Mediums, PP 322.99) for 3 laps. Finally, I did 3 laps on my own.

AI, pre-race screen
Lap 1: 172 / 168 / 164 / 175
Lap 2: 171 / 168 / 164 / 176
Lap 3: 173 / 168 / 164 / 176

Me
Lap 1: 180 / 173 / 168 / 173
Lap 2: 179 / 174 / 168 / 173
Lap 3: 180 / 174 / 168 / 174

If we take the pre-race screen AI as a baseline, the issue with Sophy is obvious (and it's not a bug where performance would drop over time): it drives reasonably well, and maybe even a little faster than humanly possible, when you're ahead (you won't notice this though, other than in the rear view mirror), and performs astonishingly bad when you're behind.

So the best racing AI ever developed has been neutered to match the behavior that has plagued the vanilla GT7 AI ever since ;-(

may be quick Race has the boost option activated ? Because it's still arcade race isn't it? I waiting to challenge Sophy in a proper lobby or custom race. Quick race is not the proper way to evaluate this IA, I think.

 

[tofuhero]

As far as I tested they apply full throttle, contrary to what happens in custom races where the AI does not apply 100% throttle.

So they are being limited by some other way. If the HP was limited they would reach the same top speed every lap so I assume is the good old boost (or in this case the opposite of boost, handicap?).

Interesting, thanks. Well what I was suggesting is that PD might be dynamically adjusting HP per lap (or sandbagging perhaps) for the AI cars. From a programming POV, I suspect it was easier to alter Sophy's car performance than to try to interfere with Sophy's AI directly and tell it to change its normal behavior (to not apply full throttle).

 

[Yard_Sale]

nobody's saying that Sophy should be impossible to beat, but there should be a way to set it up so that at least it does not artificially slow down and/or rubber band. If they want to tune the difficulty they should just tune its error rate, it's a lot more fun to take advantage if the other car makes a mistake on corner entry or exit as opposed to driving by unopposed on the straight...

I hope once they finally add Sophy to custom races it will be possible to tune it a bit better to make it as hard and/or consistent as one needs to make it an enjoyable opponent and better training for racing online (where nobody's going to lift on a straight to let you go by). If it's a CPU issue I'd rather be able to set say racing only against 4-5 "good" opponents and make that trade off to get the full performance out of it.

I was going to finally upgrade my PS4 to PS5 to get access to Sophy, but if this is the direction PD wants to go with it I guess I won't bother, as I've had enough of "chase the artificially slowing rabbit for 3 laps" type races. It's a pity though, because there doesn't seem to be any game on any platform with comparable AI to race against.

In regards to the rubber banding that’s still there, believe me that’s it’s not a deal breaker the least bit. The AI reacts very intuitively. And if anything, the rubber banding always ensures that you have someone to race against should you make ham-fisted mistakes. Not knowing your skill level (for refrence I’m low A+). Racing through the pack is like picking your way through a mid-b lobby. And first and second place give me all I can handle. The key when racing Sophy, is to race her exactly how you would a human. If you shove right through her of the way like you would normal AI, the experience will be less than satisfying.


….for you and her both, I’m sure 😳

 

[Bojador]

In regards to the rubber banding that’s still there, believe me that’s it’s not a deal breaker the least bit. The AI reacts very intuitively. And if anything, the rubber banding always ensures that you have someone to race against should you make ham-fisted mistakes. Not knowing your skill level (for refrence I’m low A+). Racing through the pack is like picking your way through a mid-b lobby. And first and second place give me all I can handle. The key when racing Sophy, is to race her exactly how you would a human. If you shove right through her of the way like you would normal AI, the experience will be less than satisfying.


….for you and her both, I’m sure 😳

You have a point but when you just breeze by them when you didn’t even get a good exit or they’re just too slow in a straight the immersion just breaks because they are artificially letting you overtake to go to the next battle.

 

[V1LR-GUE10]

ngl sophy aint it gurl...

raced a bone stock rc f at sarthe and the sophies were holding up strong and were competitive. after a lap, I overtook the c 63 s and giulia gta that flew past me earlier on the straight as if they were only using half throttle. they definitely still using the weird rubber banding crap and its getting annoying 🙄

so actually sophy is it, its just seems to be the rubber banding

 

[wdreamsmaycome]

In regards to the rubber banding that’s still there, believe me that’s it’s not a deal breaker the least bit. The AI reacts very intuitively. And if anything, the rubber banding always ensures that you have someone to race against should you make ham-fisted mistakes. Not knowing your skill level (for refrence I’m low A+). Racing through the pack is like picking your way through a mid-b lobby. And first and second place give me all I can handle. The key when racing Sophy, is to race her exactly how you would a human. If you shove right through her of the way like you would normal AI, the experience will be less than satisfying.


….for you and her both, I’m sure 😳

the issue is how the "racing through the pack" works, in lobbies if you race through the pack typically you pass due to better braking, better exit, better apexes etc. etc. you don't pass because the car in front is going much slower: this is what I mean by I wish the "detuning" of Sophy was not simply "remove HP from the car" because that is not realistic in a racing situation and does not teach anything about how to pass drivers in comparable machinery.

I wish the detuning was more along the lines of "at a x% rate give Sophy y ms lag on inputs for z ms" (causing it to randomly overshoot braking zones and/or miss the apex) so that yeah you might catch up a bit due to Sophy making a mistake, but to pass you'd have to actually do some work and not just breeze by on the straight. I have not tried Sophy 2.0 being still on ps4, but from the above it definitely makes me want to wait to upgrade to ps5 until there is some feedback that Sophy can be configured to behave in a realistic manner.

I would rather spend 20 laps trying to pass a single car but actually feel good about the pass rather than pass 20 cars in 3 laps and feel what's the point.