I've been doing a lot of experimenting with AI skill and for the majority of classes (F3 unfortunately excluded) a skill level of 95% means that I can start mid field in a 24 car race and usually end up on the podium in a 5 lap quick race. For some classes (eg GTE), taking the skill up to 96-97% makes them just about unbeatable. I remember in the good old days of the SMS forum being politely told that at skill levels > 95% the AI simulated having track specific setups to make them more competitive. Is this also a hidden "feature" of AMS2 AI? It really seems that there's a point for some classes where the skill is completely non-linear.
The game quite explicitly says that the AI drives fixed setup when under 80% skill. Was there a different layer of track-specific customization in PC2 aside from that? I'd love some actual confirmation of this, but people in the AI customization thread have suspected that they drive within a radius of 5 from where the skill slider is set. That would mean anything higher than 95 pushes the top drivers over 100, which presumably also means they simply start cheating with extra grip and such.
I'm not fully sure about the "AI using setups - situation" right now (i think there is still some stuff let to be desired, tbh., like also laptime dropoff in the race compared to player tire wear), but a few words on the AI strength situation: AI beneath 100% will use a bit more laptime de-/increase per strength %. This was mentioned also in the changelogs, when these changes were made. AI above 100% will experience a laptime de-/increase of about 2 tenths per % on a 1:30 min. laptime base on lets say GT3 and SC Brasil cars as a rough benchmark. This provides AI getting slower under 100% with more range downwards. Wrt AI cheating: AI is using different tires and drivetrain than the player, but even 120% strength AI is driving times, that are achievable by human players. (When on proper calibration, which some classes of course aren't totally on yet. Classes like GT3, GTE, SC Brasil, F3 are calibrated for example and discrepancies are more likely track/weather based there) I personally deal with 110-113% strength AI when i want to push and it's possible to win races and get top 10 with even uncompetitive cars in a class and i'm not cheating, so respective AI pace is also not off the map. The top 3-5 drivers are the ones on the correlating strength level, while the lower cars on the grid are actually slower. Please don't get me wrong here, but if GTE AI is unbeatable for you at 97%, while being mid-field/upper grid depending on sessions at 95% (Also if it is between two different classes but F3 and GTE are usually better calibrated classes, like mentioned) then you need to get in some good ol' practice (also player affinity with cars can be a decider here), or you might be indeed at a hard limit here, basically a point when few tenths decide many positions. Especially in such a series, an improvement of 0.4 seconds on a lap (sometimes these are just 2 better corners) can make 10 positions on the grid. To summarize: Yes under 100%, the strength slider is more effective to also provide lower AI pace at very low AI settings, which was subject of complaints.
So, the way I interpret this is that AI drivers with 1.00 race_skill drive at the skill indicated by the difficulty setting. Is race_skill a direct multiplier to the difficulty setting, then? Edit: Actually, that seems to be at odds with the information in the AI customization thread (which I admit I should have reread more carefully first), which says that which could be interpreted as saying the top AI actually drives at a higher strength than they were set to. I admit I haven't read all monthly updates and patch notes, but what you just said seems to also indicate that however the AI speed is adjusted, it's the same way across the whole spectrum, and they don't get special tricks past 100%. That's good to know.
Thanks for the input gents. I’m going to do a few more tests to get some hard numbers but I’ll reiterate that there are certain classes where it seems that increasing difficulty a few clicks past 95% can result in the front runners getting 1 - 1.5 sec a lap faster. This is the non-linearity I was referring to. I’m pretty sure it’s got nothing to do with how **** I am.
It's perfectly possible, that the front runners will gain more amount of time. Like mentioned, below 100% the scaling is a bit bigger, this has nothing to do with how **** anybody is, but with how the scale of strength can be used to also provide more headroom in the direction with lower percentages. Different AI drivers also have different multis to their actual strength, so they will also react differently to the scaling, of course.
