Joined: 10/28/2013
Posts: 130
Location: United States
How much of this has to do with site branding, i.e. simply not wanting the site to be associated with games like Math Blaster?
All this discussion seems absurdly legalistic if the real reason is "people like Carmen Sandiego, so it'll get views and clicks, but people view Math Blaster (and Sesame Street games, etc.) as a joke and we don't want to be associated with that, unless the run can find some way of making fun of the game".
Joined: 10/28/2013
Posts: 130
Location: United States
Great stuff. If the opening book could be dumped in human-readable form, we could look for any errors or refuted lines, knowing that Chessmaster would move (or could be made to move) into them instantaneously. The odds of finding a forced mate are very low, but it's possible nevertheless.
EDIT: Also, Saturn's line with the Philidor Countergambit is already a much better result than I would have expected. That's a perfect example of the kind of refuted opening that can get Chessmaster in trouble very quickly, but is in the opening book just because it exists.
Joined: 10/28/2013
Posts: 130
Location: United States
Warp wrote:
In this game move 3. Nd2 threw it off its book (incidentally, this move was the primary move suggested by Stockfish, so it was good.)
Huh, that's the Tarrasch Variation of the French Defense, which featured prominently in the Karpov-Korchnoi world championship matches. It's by no means an unusual move -- quite the contrary -- so it's odd that the Chessmaster opening book doesn't have anything for that, but covers the Ruy Lopez up to move 11+.
Chessmaster's Ng8-e7-g6-e7 maneuver looks pretty misguided to me; the main responses are either 3...Nf6, 3...c5, or 3...dxe4.
Joined: 10/28/2013
Posts: 130
Location: United States
Aran Jaeger wrote:
I'm not familiar with the game, but under the assumption that the AI's calculation time depends on the number of remaining pieces on the board such that one would expect shorter calculation times for it to get to a resulting turn, couldn't it possibly be a viable approach to (among the set of turns that are within its ''book'') keep doing turns that quickly reduce the number of pieces on the board with trades (to get rid of a bunch of pawns, for example), and to then break out of the AI's known set of turns after a whole lot of pieces are off the board (in the hope that the calculation times now would be significantly smaller) and to then try to win the game?
In my experience as a player of old console chess games, this usually isn't a viable strategy -- some engines take the same amount of time even when they have only one, absolutely forced move with no alternatives! -- and when it is viable, it only works once the number of pieces drops to 5 or fewer.
Joined: 10/28/2013
Posts: 130
Location: United States
Warp wrote:
I think that the suggestion of finding an opening trap (that's not in Chessmaster's opening book, or if not using the book) could be the most promising approach. It's just that I don't have enough chess opening knowledge to remember such traps by heart, and there are quite tons of them.
Is it possible to dump the opening book? If it were available in a human-readable format it might be possible to identify the worst line that Chessmaster will voluntarily play. It's unlikely that one leads to a forced mate, but not impossible. Or, if any kind of ACE exploit can be found, we could potentially get the program to believe that 1. f3 and 2. g4 is in its opening book.
Warp wrote:
Also, from some quick testing it seems that it's really hard to induce Chessmaster to select a given opening line. Its randomization does not seem to work by using the timing of user input. (For example, if I let it start with white, it will for some reason start with c4 like 95% of the time. Sometimes it will do e4, or even f4, but I can't figure out how that decision is made. Timing of keypresses seems to have very little to no effect.)
Huh, very interesting. I wonder what the source is? Consumer chess engines generally include non-deterministic elements in their algorithms to avoid playing the same game over and over again, so there must be one in there somewhere.
I strongly suspect that, as a very early release for the system, SNES Chessmaster is a barely-altered version of NES Chessmaster, so maybe the NES game would initially be an easier target to unpack (both its opening book and RNGs).
Joined: 10/28/2013
Posts: 130
Location: United States
My suspicion would be that the best bet for a quick win on the hardest settings is some sort of flaw in the opening book. We'd be looking either for an outright error (e.g. a transcription error), or for a line that we now know ends in an unexpected forced checkmate. Since opening book moves are near-instantaneous, something with mate in under 3-4 moves from the end of the book line could lead to a relatively short TAS.
(Even better would be triggering some sort of glitch that causes the CPU to misread its opening book and play arbitrary nonsense moves, of course.)
Barring that, the next-best option is likely to be (1) either a closed position in which one player builds an attack on one side -- something computers are notoriously bad at evaluating -- until checkmate is inevitable; or (2) some wild sacrificial line where, again, the CPU wolfs down material until it's too late. Here's a post that features some work in that direction by Eduard Nemeth:
https://timkr.home.xs4all.nl/chess2/honor.htm
And SNES Chessmaster is, of course, far far weaker than the engines featured in that post.
Joined: 10/28/2013
Posts: 130
Location: United States
The release notes for Bizhawk 1.12.0 seem to indicate that it's expected to run on OS X 10.9.5, but I'm getting an instant crash on startup with all 1.12.x versions under Mavericks. (1.11.x works.)
The Console log says:
"8/25/17 12:56:14.335 AM com.apple.launchd.peruser.501[152]: (org.tasvideos.BizHawk.104576[2604]) Exited with code: 1"
When I try to run the executable from the command line, I get the following, which may or may not be relevant:
Unhandled Exception:
System.MissingMethodException: Method 'Array.Empty' not found.
at MonoMac.AppKit.NSApplication.Init () <0x7d8780> in <filename>:0
at MonoMacWrapper.Program.Main (System.String[] args) <0x7d6f70> in <filename>:0
[ERROR] FATAL UNHANDLED EXCEPTION: System.MissingMethodException: Method 'Array.Empty' not found.
at MonoMac.AppKit.NSApplication.Init () <0x7d8780> in <filename>:0
at MonoMacWrapper.Program.Main (System.String[] args) <0x7d6f70> in <filename>:0
Also, do any more recent versions support the RAM Watch or RAM Search features in OS X? I tried using RAM Watch, but 1.11.9 simply crashes if you try to create a new watch, and RAM Search simply gives the same blank screen that earlier versions show for both RAM Watch and RAM Search.
I'm trying to find some way of auditing Game Boy RAM under OS X, but maybe it can't be done.
Joined: 10/28/2013
Posts: 130
Location: United States
I'd like to see a California Games TAS on the Genesis, partly to find out whether or not successfully completing the game's "Compete in All Events" mode offers any kind of additional ending screen/sequence beyond what you get with the individual events.
Joined: 10/28/2013
Posts: 130
Location: United States
Still hoping to see a Private Eye TAS someday -- as well as Porky's (which is absolutely begging for a TAS, since it can be beaten in something like 5-10 seconds) and Chuck Norris Superkicks.
Solaris would be great too, as well as Starmaster.
Joined: 10/28/2013
Posts: 130
Location: United States
In the NES version, setting up the board, taking moves back, or changing sides mid-game would automatically mean that you wouldn't get the "CONGRATULATIONS!" screen when you won.
I guess the SNES version has no such screen; I couldn't remember one way or the other, but now that I check old screenshots of mine, it seems to be the case. So pretty much anything that delivers checkmate will yield the same results (another piece of evidence that the SNES port is really quite shitty, and the NES game is the better one).
Still, I wish people would find a way to break the game on its toughest settings, rather than put it in a mode where the CPU deliberately handicaps itself (by simply playing the first move that comes up in its search tree) and call that "beating" the game. At least Newcomer 1 was the default in the NES version; here, you have to select it, thus dumbing the game down from its defaults.
It'd be much more interesting to find a way to trick the CPU into poor play, to find a hole in the program's opening book that leads to forced mate (has anyone dumped the opening book?), or to somehow corrupt memory so the Level 13 AI really does believe 2...Ke7 is the best move after 1. e4 e5 2. Qh5. How does Chessmaster respond to multitap or expansion port inputs?
Joined: 10/28/2013
Posts: 130
Location: United States
^Thanks for posting, ktwo. I don't have the chops to help, but I hope someone steps up to assist you. I have to believe all the bugs, randomness, and other weirdness in this game offer some way in to an impressive TAS, and perhaps exploits that can be used by RTS as well.
Joined: 10/28/2013
Posts: 130
Location: United States
Are you looking at jzintv for guidance at all? I'm surprised that a whole new emulator core would be developed from scratch when an open-source core already exists in the form of jzintv (which is highly accurate in most respects & the standard emulator used by Intellivision developers), but I'd imagine you have a good reason.
Joined: 10/28/2013
Posts: 130
Location: United States
4-player runs of this game are one of my absolute favorite forms of TAS. It's like watching a cross between ballet, video games, and four super-high-end assassins carrying out a hit.
Joined: 10/28/2013
Posts: 130
Location: United States
I'm grateful for this run either way since, following the strategy used here, I was able to beat Overlord quite easily -- which is satisfying for a game that's allegedly among the hardest on the NES.
The only tricky part is the tank battle at the end, but I just drove in circles, shooting once per revolution. Since two of the turrets only shoot when you do, and the third fires a spray with a short range, you can take them out one at a time with little difficulty. A sub-five minute win is entirely possible in real time; mine was around 7 minutes.
Joined: 10/28/2013
Posts: 130
Location: United States
I'd suggest going to AtariAge and talking to intvnut, the author of the jzintv emulator (the most accurate of the bunch). I'd imagine he'd be happy to add TAS capabilities if he has the time -- and if they're not already present in some nascent form.
I love the idea of Intellivision TAS, BTW. :)
Joined: 10/28/2013
Posts: 130
Location: United States
jlun2 wrote:
goldenband wrote:
If a BIOS-level vulnerability were found in a console like the 3DO or ColecoVision (both of which make heavy use of BIOS calls), and it allowed for ACE in all games, how would it be handled under the site rules?
Well, that's certainly the inspiration, but I'm talking about something where a console's entire library could potentially be vulnerable to the exact same attack. My understanding is that the PCM bug only affects games that use the PCM channel and have a specific type of controller-reading routine.
On the 3DO I'm not aware of any retail release that wasn't sandboxed inside the 3DO BIOS, so a (theoretical) vulnerability in the BIOS would potentially allow for every game in the console's library to be completed in the exact same way: ACE, jump to credits or equivalent, done.
(I'd think the same would be true if a vulnerability were somehow found in the startup splash screens for Game Boy, Neo Geo, Genesis w/TMSS, etc. Obviously that's extremely unlikely, but...)
