I prefer using my own model for privacy reasons. A crude summary is fine, I don't need GPT4 level quality. My PC has the capacity to run llama 2; I need a model to summarize my personal text documents and am looking for one optimized for this purpose. I prefer ubuntu over windows
Grand Prix Legends demo (free but not open source) is one of the sim racing games in the package. It just turned 25 years old last month, but is still played online by hundreds of fans: https://www.youtube.com/watch?v=I5OqIOIhlyA
My sentiments exactly! It is a tribute to the wine developers that these 20 year old games run as well (or better) on linux and modern hardware than on the ancient PC's and MS-Windows versions they were designed for. One recommendation - with each game try wine-6 first (./wine_default.sh), if anything is not quite right use wine-7 (./wine_experimental.sh). That should be sufficient to run all games; use the oldest wine version that works. If there's still a problem go to winehq.org and try wine-8 (winehq-stable). (You can also run most of the included game exe's on Windows 10, emulating the appropriate old Windows version. If using Windows, the linux install script serves as handy documentation for the install, showing which patches to apply, what directories to copy, what settings to change, etc.)
The racing sims include dozens of cars and tracks, Combat flight sims, which simulate entire wars, include terrain and hundreds of vehicles and munitions. There are dozens of games in this package. There's no need to install them all, but if you did, you'd need about 210GB for a single installation, more if you don't delete intermediate installation files
The car and flight sims are engineering simulations - differential equation solvers employing multiple degrees of freedom, flight models and tire models validated vs. data, etc. Game manuals, which are hundreds of pages long, describe how to, e.g., set your differential coast and power ramp angles to tune racecar responsiveness, and how to decide whether a one-circle or two-circle dogfight is more advantageous based on the specific power diagrams of your aircraft and your AI opponent's. These games have staying power because they are clearly labors of love, where accurate physics was top priority in game development. Dave Kaemmer, a top sim racing developer, summed it up this way: "We don’t care if it’s fun or not. We care if it’s real"
There's a python script that makes the organization more clear. Once you've installed the games package, run "./launcher.py" at the ubuntu command line to see a text menu "Top Level Menu" which launches the games.
So the original Star Wars, which has spawned so many copycat movies, is itself a copycat movie? Let's see. The formula seems to be: take two successful old movies, such as Battle of Midway and Hidden Fortress, and create a mashup set in a different genre to make it seem new. Has someone created a machine learning application that applies this mashup/genre switch on old movie scripts to make new ones?
The link to the tar file containing windows binaries that run under linux/wine is in the Readme.md file of the repository linked to by the hacker news submit here: https://github.com/bencaddigan/esports-for-engineers
For the entrepreneurs among us, the business story here is that these fantastically detailed sims were expensive to make. The three 20-year-old sims at the github link, Grand Prix Legends (GPL), Mig Alley (MA) and Falcon (FF) were all bet-your-company high concept projects that expected a lot from their customers in 1999.
MA did OK in the market but that only encouraged it's creators to make a follow-on product that tanked their company. GPL and FF were commercial flops, and the companies that made them soon disbanded. Nevertheless, GPL and FF evolved into what are currently best-in-class sim products. The high ambitions and technical wizardry of all three sims inspired dedicated community groups that added content to the sims for years following release. The versions of the sims on the github page include these community-generated add-ons.
The FF project was a notorious business mess that somehow propelled itself through years of headwinds with a kind of prodigious creativity. Both the mess and the creativity are on display in the 1000 page volunteer-created FF manual on github.
Debates about which sim is more accurate bring up a philosophical question about sim design. My take is that designing a realistic simulation is like solving a physics problem, in that it's impossible to capture all of reality with a mathematical model, so the challenge is to first decide what you want to model accurately and what you can ignore. Doing that well requires insight. Case in point - although Rowan got overwhelmed by 1C in the flight sim marketplace, I would argue that Mig Alley is better than the IL2 product that covers the same material because MA is more accurate where it matters. IL2 has more realistic cockpits but Mig Alley does a much better job of modeling the physics of the Mig/Sabre asymmetry. That asymmetry is what makes MA dogfighting so unique. And once you've mastered that one-on-one duel, you find that the many vs many contest is completely different and involves little extended dogfighting- also true-to-life according to the history books I've read.
I agree completely. The other two games in esports-for-engineers are open source, while the GPL demo is free but not open source. (I didn't have enough space to explain this distinction in my one-line hacker news post.)
iRacing is certainly more accurate than GPL, but that's an apples to oranges comparison because these sims have different goals, and both are great at what they do. If you are preparing to compete in real-world car racing, by all means sign up for iRacing (after reading the excellent and funny book "You Suck at Racing"). If you are an engineer, and like to learn about car physics by adjusting the car settings in GPL Race Engineer and then seeing the results in GPL Replay analyzer (both included at the github link), then GPL is a fine choice. There's a lot to be said for a simple, elegant game design that provides an immersive experience; By that measure I submit that all three games on the esports-for-engineers github page deliver.