ChatGPT as a sim racing driver coach?

BMW in iRacing.jpg
The term "Artificial Intelligence" is not new, especially for a simracer. We often practice or just have fun by "racing against the AI".

Outside our hobby you have surely read (about) books or seen movies where AI plays a role, sometimes even as the protagonist or antagonist of the story itself. HAL, C3PO, The MCP, Replicants, Terminators, Data, Chappy, Cortana, Jarvis... you get the idea. Man-made constructs that behave and speak like people. But AI is not science fiction anymore, it has not been for a while actually. However, it has just recently entered a new phase where chatbots (a digital entity that can simulate textual conversations) use complex language models, large databases of information and reinforcement (machine) learning techniques to recreate responses on queries that simulate those of (well informed) human beings. In simple words: they respond to questions the way a person would.

ChatGPT​

ChatGPT is one of those chatbots that, after being released for free by OpenAI at the end of 2022, has been the talk of the internet. Particularly because ChatGPT can create large sets of text that is so correct, coherent and nuanced that can pass for words put together by a real person. News and especially universities have been taking notice, trying to react and anticipate what this technology means for different areas of work and education. ChatGPT is very interesting for research and people have used it to do their (home)work, writing things from a simple article to a graduation thesis. Schools across the world are currently trying to understand how to deal best with this technology as the impact it has can be tremendous for human learning.

Positive or Negative

But the power of any technology is as positive (or negative) as the intentions and capabilities of the user. And one very interesting case can be seen in the video below, where an attempt is made to use ChatGPT as a driving coach for iRacing. This experiment is very interesting as it showcases how this technology works and can be implemented to teach a real person the basics of a specific discipline... such as sim racing. Enjoy!

About author
Dzul
Raul Martinez. Father of one, design teacher of many. Skeptical technophile, whisky philosopher and Simracer since the times of Pitstop II.

Comments

Fascinating video and while the AI is amazing it does give you possible information that is convincingly wrong and this to a noob such as myself could really put me on the wrong path to improvement.:(

Kudos to Cameron as he already knew the correct answers and was able to pull up the AI for some incorrect information, so this technolgy is possibly for experienced users that already have some knowledge of the subject matter that the AI is producing (personal opinion)....amazing technology :thumbsup:
Indeed, a human being is barely able to distinguish the right information from the wrong ones. It depends on his own knowlegde of the subject, on his ability to search for information about the sources to evaluate their credibility in general and also about this specific subject (the so-called "intellectuals" giving their opinions about subjects they can't understand all over the net and in videos are a real danger, they can, exactly like this AI, state stupid things with the most convincing power possible - more stupid, more convincing, that's the rule - :D ). It takes time and energy, and some skills to make the research.

All of this process can't be done by an AI, unless all the details of our life, as human beings, are recorded and readable by this AI. Well, considering people put even photos of the food they eat on social media, it will may be a truth in a few years... It is already possible for most of high level specialists to find what diplomas and experiences they have, if they published something on the internet, or are quoted, they most probably are on linkedin. Their school too. Schools are ranked each year (although these ranks are just schools paying more than others) and the ranks are probably on the net. So an AI could make a mix about all the existing ranks from all over the world, and get an idea of the academic level of this person (although diploma is not intelligence, let's not forget it, that's why it is necessary to find other information). If he is not an expert on the subject, the AI could forget his opinion by considering his low credibility (or giving him a low grade to later make an average with other people). If he is a specialist of the subject, this person may have written articles or books which may be published on the net (consulting pdf libraries, can the AI do it?). If these texts were theories or statement, have they been refuted some time later? it seems this AI could do that. The same with jobs, personal life (morality matters when evaluating credibility, but there is no sure rule ; many people have been served prison sentences, another info that could be on the net, without being immoral, or being just victim of a repressive country).

End of the example, but this would be maybe not more than 0.0001% of the research to just check one information before delivering it (giving the probably less wrong piece of info is more accurate), and it supposes to have every piece of information available on everybody over the planet at the disposition of our friendly AI. Ok...
AIs can learn by themselves, but they just learn from what humans give them.
By chance, human judgment is still better, although AI is an amazing and fascinating tool, but it is just a tool.
 
It seems to me a totally unnecessary waste of energy and time, I see it as useless as an experiment, the green stripe that the noobs put on the ground that turns red when you have to brake does the same thing as this AI.

If the objective is for an AI to effectively train a human, that is currently unfeasible for an AI for a long time, it involves many factors with a huge number of variables, each factor, without forgetting the perception of reality that this never it will have an ai.

ChatGPT can seem incredible the first time you use it, then you realize how limited an AI is and how far it is still from being a tool to replace the human.
I'll just leave this here. I asked for GPT to reply to your quote.

The statement highlights some of the limitations of AI technology and language models like ChatGPT. It is true that AI models still have limitations when it comes to understanding and replicating human intelligence, and that they may not be able to fully replace human capabilities in the near future. However, AI technology is constantly evolving and improving, and the potential applications of language models like ChatGPT are numerous and varied. These models can still be valuable tools in a wide range of industries and can help to enhance human capabilities. While they are still far from being a full replacement for human intelligence, AI models like ChatGPT have the potential to significantly improve many aspects of our lives.
 
Hey guys, just found this thread by accident. I implemented exactly that including a data integration with data from the current session. Is part of my complete software suite with four different AI - based Assistants. You can find everything here:

 
Last edited:
Hey guys, just found this thread by accident. I implemented exactly that including a data integration with data from the current session. Is part of my complete software suite with four different AI - based Assistants. You can find everything here:

This looks amazing.
 
Premium
Hey guys, just found this thread by accident. I implemented exactly that including a data integration with data from the current session. Is part of my complete software suite with four different AI - based Assistants. You can find everything here:


So, this is a pretty cool implementation of an AI angineer that talks to you during the race


Open the video description for clickable timestamps where you can hear the AI talk

1702231995203.png


It's a little clunky but we're starting to see what Bell's GT Revival could accomplish and maybe do it even better with their resources.
 
----

It's a little clunky but we're starting to see what Bell's GT Revival could accomplish and maybe do it even better with their resources.

This example demonstrates Assistants which are implemented using an old-fashioned, rule-based approach. Still very capable for deterministic stuff like strategy calculation and pitstop preparation, but not as amazing in the communication as an GPT-based Chatbot. There is a fourth Assistant in "my" crew, the Driving Coach, which is using an LLM (actually you can use any LLM, even ChatGPT-4 by configuration) to assist you in questions of handling issues, car setup, and other driving related stuff. This guy (or girl, if you want) is far away from being clunky.
 
Last edited:

Latest News

Article information

Author
Dzul
Article read time
2 min read
Views
16,706
Comments
46
Last update

What's needed for simracing in 2024?

  • More games, period

  • Better graphics/visuals

  • Advanced physics and handling

  • More cars and tracks

  • AI improvements

  • AI engineering

  • Cross-platform play

  • New game Modes

  • Other, post your idea


Results are only viewable after voting.
Back
Top