I don't think there's any way that exploit can be made to be entertaining. It also takes a noticeable amount of gametime as well as realtime; you can only fit 52 items in open inventory, and you can't rename items that aren't in open inventory, so you waste a bunch of time moving items around.
I was going to say "blessed figurine of an Archon" (which you don't pick up, just walk over) but I think "cursed figurine of an Archon" (which you DO pick up, and manipulate to transform at the right time, substituting for an egg) would be better.
Other suggestions:
stack of scrolls of punishment in order to get both 1. an iron chain and 2. a super-heavy iron ball.
A saddle, tin opener, or other annoyingly rare tool like a grappling hook.
A cursed bag of holding in which you carry a bunch of stuff and completely don't care about the chance of it eating things.
Silver saber, shirt. A towel if you're not a hoopy frood yet.
Demogorgon.
Wand of Nothing (0:8)
Slime Mold
Decided how to kill(or not-kill) Vlad yet?
A figurine of a Woodchuck is completely out of the question without wishing, yes?
Don't restart. Finish the current version and then start on a v2. It takes basically the same amount of time, but you learn more things from the endgame that you can use in your next version; you might get more feedback from the outside that reveals further improvements by publishing; publication renews interest and might inspire others to try to TAS the game; and this way we reduce the risk of nothing being published at all due to redoing burning you out.
Publish early and often.
Moderator, Senior Ambassador, Experienced player
(907)
Joined: 9/14/2008
Posts: 1014
I'm in favor of not restarting, but for very different reasons. One thing to keep in mind about our stated goal of lowest in-game turns is that anything that happens before turn 2,000 is malleable - while we don't want them to look sloppy, we have a lot more freedom than we do on the turns after 2,000 until the end of the run. What makes this look sloppy in ais523's eyes is that we were standing right next to two leashes (although we did not know that at the time) and we spent more time than needed reading spellbooks because we should have been a faster polyform (a reasonable statement).
Having said that, I seem to recall that we were reading the spellbooks to kill time while burning through the RNG so that we had more favorable results after that point, and even if we weren't actively planning on that at the time it's a very plausible excuse. :) I'd also say that while we know now that there were two leashes there the entire time it's not something that I think any audience would take us to task over - whether we mention it or not is a debate for later, but simply picking up two leashes (either those two or others) in some interesting way later could be just as entertaining. As it stands now we'll likely have "turns to burn" prior to turn 2,000 anyway because we were so careful with our turns earlier so I suspect this will all be a moot point.
As far as starting over / someone else doing a TAS, I suspect that this particular effort will never be attempted by anyone ever again. If we do the turns after 2,000 correctly, it will not be possible to beat the same category unless a substantial new trick is discovered, which is unlikely given the time ais523 and others have spent scouring the source code (a rare privilege not often experienced when working on a speedrun). I would definitely encourage others to try if they want to, of course; I have released the framework we're using as open source alongside ais523's modifications to NetHack (nethack-tas-tools) but I don't see it as being very likely. It's perhaps more probable if a late-game strategy is discovered and the input up to that point is preserved, I suppose.
At any rate, those are my thoughts - I definitely think we should keep going. :)
A.C.
******
I'm more of the opinion that the goal of a TAS is entertainment, and just because there's a frame rule coming up isn't an excuse for looking sloppy. (And indeed, I think it's unlikely that anyone else will take up TASing this game, although I'd enjoy it if omeone did.)
I guess the real question is to do with how much needs to be redone, which is unclear. NetHack's sync-stable enough that bits of existing input sync. If we have to redo the entire Castle, it probably isn't worth it. If that syncs, redoing gold duplication wouldn't be so bad.
I think we've tentatively decided to continue onwards rather than redo an earlier section; it's sloppy, which disappoints me, but redoing it would be a lot of work and the time lost just runs up against a framerule anyway.
We've made some progress (not yet documented); we grabbed a couple of wands of wishing from a shop (confusing the shopkeeper into letting us have them for free, no "theft" involved), and are moving on to boulder collection. We're planning to do the first level of Sokoban halfway for the scrolls of earth, and to give people some idea of what it looks like under TAS conditions but leaving before it overstays its welcome.
The project was stalled on me writing a Sokoban solver that could handle all the methods of cheating that a TAS has available, but I was thinking about the problem a while ago, and think it may be possible to come up with an optimal strategy by hand, proving it's optimal rather than needing a computer search. I just need to stare at the map a bit in order to make sure it works (going off to do that now), and get some implausible levels of luck, but it's a TAS, you can always expect those :-)
I think this solution works! Livestreaming it right now over on termcast (telnet termcast.org).
EDIT: Gah. First I thought we'd have to backtrack out of Sokoban, so I created a solution that got us back to the door. Then I thought we wouldn't, so I made an alternative solution, also adding in jumping (which I forgot the first time). Then it turns out we can't leave except via the front entrance anyway, so backtracking it is. Time to combine the two solutions…
EDIT 2: Here's the map after the obvious initial moves:
We're the H. We need to pick up the = and both ?, then return to the >; the ^ block our path unless we push or teleport a 0 onto them. One of the 0 is in place for pushing already; for the other three, we use teleports. We can jump up to three squares orthogonally, or two in any direction. The / is an irrelevant wand of undead turning (0:4) (somebody's bound to ask).
The first teleport needs a seed of 104137 to succeed (or many other odd numbers around there; there's massive jitter), and I don't think it matters which boulder we teleport. The other two teleports are somewhat over 700 seeds after that, and a little over 500 seeds after that, with the exact details depending on our actions (but those values are close enough to quickly find them by hand).
EDIT 3: Turns out the second solution is the fastest anyway. Due to the awkward jump distances we end up with, it's best to pick up the ring on the way there anyway, and just backtrack back in a comparatively boring manner.
EDIT 4: I declare the raid on Sokoban complete! Our earnings: two scrolls of earth, a ring of conflict, and something like 120 rocks. Not sure where to go next. (Also, we have our luck at its lower cap of -7 right now due to all the cheating; we might need to go fix that. Luckily, fixing luck issues is not hard. There may be something we can do with the negative luck before it's fixed, though.)
Also considered: punishment dragging, but it's slower than jumping except when in a 3-actions-per-turn polyform, so I discarded the idea.
I'm really busy at the moment. dwangoAC's reacted by making improvements to the tools we're using; we don't have much/any more progress on the actual run.
Moderator, Senior Ambassador, Experienced player
(907)
Joined: 9/14/2008
Posts: 1014
Hi, thanks for your interest! As ais523 noted, he's been extremely busy this fall and I'm actively preparing for AGDQ 2015 and have not made any progress on the newer framework method I've been pondering (using CRIU). My hope is that I will be able to finish the framework early next year and that we'll be able to resume meaningful progress in time to submit the run to the site before the deadline where the game we chose in the future is, er, no longer in the future. :)
I also have dreams in my head of trying to do NetHack at a future AGDQ or SGDQ event in a graveyard shift, where we spend two hours or so and walk through explaining all of the complex things going on but fast-forward through the boring parts, explaining in enough detail that people unfamiliar with the game can understand what's happening. So far I've received extremely divided opinions on whether this is a good idea, with the general response to the idea being either one of utter elation or utter repulsion and not much middle ground. I'll have to ensure that whatever we pitch meets the needs of as large an audience as possible.
Thanks again for the interest and I hope to have something useful to put in here sometime in the next couple of months!
I don't personally think NetHack would be a good fit for a GDQ. However, if it somehow happens anyway, I'll do my best to make it entertaining.
And yes, I haven't been working on the run. I'm incredibly busy with non-TASVideos-related things at the moment.
This is actually a really effective way to identify items in real time play, though a bot is of course flawless at doing it, whereas a player will usually need a helper program.
Yep, this is actually the normal way to ID scrolls (and to a lesser extent other items). The difference is that humans need to keep notes, bots can just memorize the prices.
This is not meant as a correction/criticism, but merely an interesting philosophical observation that struck me reading the last two posts: it's kind of odd that it felt natural for you to distinguish between the actions "human takes notes" and "bot memorizes" in that way, since of course these actions involve the exact same steps - store the information somewhere, then retrieve it when necessary. Usually human notes will even be stored in the same medium as a bot's - in RAM/on disk. And yet, when a human does it, it feels like an extra thing they have to do, but for a bot, it's "just memorizing".
I more or less understand why this happened - we have some loose model of gameplay as by default only including interaction between the game and the player, and not the player and outside objects. For the bot, (some of) RAM is included in our perception of the bot's self, so it doesn't seem to violate the model, but for the human it does.
Sorry for the random tangent!
A warb degombs the brangy. Your gitch zanks and leils the warb.