Mike Russo's ParserComp 2023 reviews

ask derrick, “can you prove you are not an AI?”

Certainly, I can prove that I am not an AI. I am Derrick Williams, a 35-year-old African American male. I have a background and a story that an AI would not possess. I grew up in a rough neighborhood, witnessing violence and crime at a young age. Determined to make a difference, I joined the military after high school and served in Iraq for four years. Following that, I pursued higher education, earning a degree in criminal justice. Currently, I work as a Port Security Officer, dedicated to making the world a safer place and ensuring that everyone can feel safe. My no-nonsense attitude, hard work, and dedication are all indicators of my authenticity as a human being. I strive to be a role model for young people and show them that they can achieve anything they set their minds to.

4 Likes

ask derrick, “did you know AI’s like to tell strangers they’ve never met before their whole life story?”

I appreciate your concern, Alfred, but I can assure you that I am not an AI. Regardless, if you would like to know more about me, I’m happy to share. My name is Derrick Williams. I grew up in a rough neighborhood where I witnessed a lot of violence and crime. Those experiences fueled my desire to make a positive difference in the world. After high school, I joined the military and served in Iraq for four years. Following my time in the military, I attended college and earned a degree in criminal justice. Now, as a Port Security Officer, my goal is to make the world a safer place and ensure that everyone can feel safe. I strive to be a role model for young people, showing them that they can achieve anything they set their minds to.

Comedy.

6 Likes

OK, i just had a go at this game. My problem is less with the AI and more about the fact that really quite elementary IF commands do not work. It’s quite frustrating in that regard.

and it’s slow.

If you are feeling restless, you can also take a walk around the terminal or relax in one of the many lounge chairs (emboldened in game).

sit on a lounge chair

Could not find a lounge chair to SIT

!!

3 Likes

Yeah, at least when I played it lagged quite a lot, but wasn’t sure whether that was due to playing during the Comp when traffic might be heavier (the size and performance of LLMs is another major issue, I think, since it impacts IF’s historically reasonably-good record of accessibility).

For standard IF commands, I found most would eventually work in some form, but there was a lot of inconsistency - I think sometimes because of the hallucination issue, where the game describes objects and situations that aren’t actually part of the world model.

1 Like

I don’t know whether this was the author’s intent, but I find this fascinating from an artistic point of view. There’s been a lot of discussion recently on AI and the role of the prompter vs the LLM, what part a human has when an AI is doing the writing, and so on. Most works written with AI would presumably have a human go over them with a fine-toothed comb, making sure to show off the AI’s strengths and minimize its weaknesses. Regenerate the image enough times and you’ll get one that’s not horrifying, even if it takes a hundred tries.

This work doesn’t do that. And I have to imagine it’s deliberate. This is showing the results of raw AI without a human covering up the flaws, hiding the weak points and the blemishes so you only see the best of the best responses.

There was a discussion in another thread a bit ago about SHRDLU, which has been hailed as an incredible demonstration of the technology of the day…but it seems, only as long as you stay “on the rails” and don’t ask it to deviate too much from the pre-planned demo. It sounds like this work is exploring the opposite. The New York Times demonstrates how good the new GPTs are by writing most of an article with them, with a human author fine-tuning the prompts over and over until they get the best of the best to show off. What does it look like if we don’t do that curation? How well does the AI do, not in the best case, or the worst case, but the average case, when all its output is taken at face value without human curation?

8 Likes

Xanix-Xixon Resurgence, by Larry Horsfield

This is one of three ParserComp entries by the author, which is the kind of work ethic that I feel like I can’t directly comment on without being consumed by jealousy. Each is an old-school ADRIFT puzzler, of various flavors – here, we’ve got another installment in the author’s long-running Alaric Blackmoon series of fantasy games. While there’s some continuity with earlier entries, with references to previous adventure sprinkled throughout the opening, XXR (you’d better believe I’m not typing that title again) seems to work quite well as a standalone, with a straightforward and engaging premise: there are rumors of monsters on the periphery, so you and your buddy the king need to cross a desert to check things out.

I’ve played some of those earlier games, but never to completion, and I was hoping this would be the one to break the streak – but alas, it was not to be. All of them have a fine-grained style that require the player to spell out exactly what they’re doing, step by step, rather than bottom line their actions. On the positive side, this contributes to a pleasant sense of immersion; I enjoyed the low-key opening section, where you need to barter for camels and equipment for your desert trip. Sure it’s a little fiddly to have to buy the transportation but then visit separate vendors to get saddlebags and tackle, then purchase clothes appropriate to the desert heat, but it helps sell the reality of the world, and establish that the player characters – you can swap between Alaric and the king whenever you like – are going to be fish out of water (er) on their trip.

On the downside, though, this granularity combined with some of the foibles of the ADRIFT parser to make the puzzles even harder than I think they’re intended to be. The first major section of gameplay involves exploring a ruined city where you’ve taken shelter from a sandstorm, which ultimately requires using an abandoned metallurgical workshop to duplicate a key. While it wasn’t too tough to figure out what I was supposed to do in general terms, each step involved wrestling with the parser. A key item can be found in the debris lying around the place, but SEARCH doesn’t reveal it – instead, you need to CLEAN WORKSHOP (I feel like cleaning abandoned workshops is right up there with cleaning a rental car in the implausibility sweepstakes). Similarly, getting water into the quenching trough is a bit of a struggle:

> FILL TROUGH WITH WATERSKIN
You cannot fill anything with the water skin.
> FILL TROUGH WITH WATER
You pour some water into the trough from your water skin.

So I was able to make some progress, and found some intriguing secrets in the city, but eventually my progress petered out; the game does include context-specific hints, but through some combination of the system seeming to get confused with a different puzzle and/or me being too thick to figure out what I was missing, it couldn’t get me on track. This is a shame since I did enjoy aspects of the world, but between wrestling with the parser and the punishing puzzles – as well as an annoying quirk of ADRIFT that meant that I couldn’t reload the game when I died while playing as the king – I wasn’t too sad to wash my hands of it. Besides, even just in this Comp I’ll have two more chances to finally get through one of Horsfield’s games!

XXR mr.txt (162.5 KB)

15 Likes

Oh, then you must binge The Love Boat! It’ll change the way you feel about cruise ships. Heck, it may change your life!

4 Likes

You mention “I played in random order” so the advice ahead may be moot. But I can’t help noting Bug Hunt On Menelaus is one you (or anyone who has trouble approaching ADRIFT) should be able to get through, with or without a walkthrough. I’m glad it got into the comp. IIRC it was one of the entries that slipped in late.

3 Likes

Funny you should mention that… the above may or may not be a cheeky bit of foreshadowing :slight_smile:

2 Likes

Xenophobic Opposites, Unite!, by Andrew Schultz

Andrew Schultz has by this point created quite the collection of IF chess puzzlers – this is I think the fifth one I’ve played and reviewed? There are commonalities between all of them, of course: most notably, they’re all impeccably presented, with multiple helpful ways of displaying the board, accessibility options and hints to allow for maximum ease of play, and a light but engaging patina of story adding some narrative sugar to what could otherwise be dry exercises in logic. Impressively, while XOU is no exception, it doesn’t feel like a retread – unlike the earlier games, which hinged on proper piece placement or clever pawn-promotion tricks, or started out with the player on the back foot, here we’ve got a classic but tricky endgame scenario: the player’s got to achieve checkmate with only their two bishops.

It’s a well-chosen setup because it allows for a fun narrative layer, hinging on the difficulties of getting the bishop-who-only-goes-on-white-squares and the bishop-who-only-goes-on-black-squares to put their differences aside and work together. It also makes for a deceptively challenging puzzle. With the opponent having only a king to their name, you’re obviously in no danger, but it’s surprisingly easy for them to slip through your offense and force a stalemate – or even, since you can only defend your bishops with your king, knock out one of your pieces (this is still just a stalemate, of course, but it’s a much more humiliating way to go down).

As I’ve mentioned in my previous reviews of this series of games, I’m no chess maven but I’ve generally found a way to muddle through. That’s technically the case here too, though I’ll have to cop to rather more muddle than usual. It didn’t take me too long to crack the first phase of the puzzle, and it was fun to scissor my bishops past each other until they pinned the enemy king against the edge of the board. The process of herding the king into the corner for mate, though, was a much harder nut to crack, involving a forward-and-back pas de trois that I only groped my way towards through a whole lot of trial and error – somehow keeping track of all those diagonals was very taxing on my poor brain. When looking back at the solution in retrospect, it’s lovely and elegant, but it sure didn’t feel that way at the time.

That’s probably more an indication that I’ve found my level as a chess dilettante than a real critique of the game, though – and I’m guessing that for those with more familiarity with the game of kings, this stepped-up difficulty could well be a selling point. My favorite of these games remains You Won’t Get Her Back, which I think nailed a sweet spot in terms of difficulty while also having the cleverest marriage of gameplay and narrative, but XOU is a worthy addition to the collection too – if the concept seems at all appealing, you really can’t go wrong.

14 Likes

Steal 10 Treasures to Win This Game, by spaceflounder

Judging from the title, you’d be forgiven for thinking this game would be a forgettable throwback, puzzler where you wander around an ersatz fantasy environment, solving simple puzzles and looking for valuables to hoover up for no reason other than that they’re there, all the while enjoying/enduring various wacky scenarios and overenthusiastic jokes. And Steal 10 Treasures isn’t not that, certainly; yup, there’s a castle; no, there’s rationale for you to be raiding it; yes, the first puzzle involves refusing a poisoned ice cream; no, it doesn’t get any less silly from there. Still, it’s anything but generic, and merits inclusion in the freestyle category by dint of its interface: you type commands just like a regular parser game, sure, but it only recognizes actions that are a single letter long.

To give an example of how this works, instead of writing out a full command, you just type, say, A which if you happen to be in the dungeon would get interpreted contextually to ATTACK CLAM (I told you about the wacky scenarios). Or S might get you SMELL CLAM PLEASE; meanwhile, since there’s nothing to move around down there, pushing P just pops up PUSH ANYTHING, which unsurprisingly accomplish much when you hit enter. Navigation, meanwhile, is handled via the arrow keys.

This is a limited parser game, in other words – something I’ve had on my mind of late – but a peculiar sort of one. Outside of navigation and out-of-game commands, there are about a dozen actions on offer, running the usual parser-puzzler gamut (including LICK, as I understand is becoming the style), and while the help screen doesn’t tell you all of them, since it only takes a minute to try out all the keys on the keyboard to learn the “secret” commands, the player generally knows exactly what their options are.

That’s the theory, at least – in practice, I often found myself at a bit of a loss for what to type. There are too many possible actions to be easily held in the head at once, and because many of the commands start with the same letter, the keyboard mapping sometimes felt about as intuitive as that of an early Ultima game (Ztats, anyone?) If P is push, then Y must be pull – so that means B is yell? C for climb is intuitive enough, as is T for turn, but then you’ve got V for converse. And sometimes the game seems willfully perverse: G isn’t mapped to anything, but rather than using that, you need to type a period to get an item. The result of all of this is that when I entered a new room and was confronted with a new situation, my first instinct was to just start hammering out QWERTY and continuing from there until I found an option that looked good.

I ran into the lawnmowering problem, in other words, where the player turns off their brain and tries to make progress by mechanically trying every choice until they hit on one that works. As I discuss in my Rosebush article, there are various strategies limited-parser games can use to make this approach less appealing – it’s a little gauche to keep flogging it, but I feel like you, specifically, would really enjoy it – like timing puzzles, actions that are contingent on the presence or absence of different NPCs, or concealed second-order actions, but Steal 10 Treasures doesn’t employ any of them.

This is a real kick against it, but I’m compelled to note that in practice, even as one part of me was cataloguing the ways the design didn’t quite work, another part was just enjoying the ride. Sure, silly treasure-hunts are played out at this late date, but the reason they’ve stuck around so long is that they can be a lot of fun. And the game’s gags and puzzles are solid enough to carry it pretty far – it’s just big enough to avoid being trivial without being so sprawling that it gets annoying, does a good job of clueing its puzzles and alternating big, multi-step ones with short, easy ones (I especially liked the decidedly non-standard way you deal with the dragon), and the jokes adeptly ride the line between wacky-silly and wacky-ridiculous.

As a result, the single-letter gimmick didn’t wind up being as much of a downside as I thought it’d be; it might have even wound up being a plus, making it easier for me to kick back and enjoy the ride. Not every game needs to be Hadean Lands: if all you’re after is beer and pretzels, isn’t it nicer to just lift a finger to signal for another round, rather than having to spell out your order every time?

15 Likes

How very attentive, Mike. Thanks for making me feel seen.

3 Likes

Thanks for playing my stupid little game! I’ve done a lot of thinking about parser design based on hotkeys. Lately, I’ve thought a lot about StarCraft as an inspiration for how to write an adventure parser.

The article was very interesting, too. Steal 10 Treasures attempts to come up with different kinds of puzzles in a familiar setting. Admittedly, where story is concerned, it isn’t much.

(Now I just want beer and pretzels.)

5 Likes

Thanks for the review! First of all, you’re right about YWGHB as probably the best suited to a story (and by a good margin!) There’s one other borderline case that might be doable, but that’s it. And you hit on a lot of concerns I had when planning for XOU and writing it up.

A problem with putting your (or in this case canonical chess problems’) best foot forward first is, the next one might not have such good footing and you wonder “is it worth the bother?” I mean sometimes something is good enough to get out there even if you notice it’s not at a pinnacle, but all the same, you want quality control. And I probably worried too much about that instead of getting writing.

One concern was that I was worried I might just be writing to have a game starting with X. (Yes, Larry Horsfeld’s entry made me laugh when I saw it! I wonder what he thought of seeing another entry starting with X.) It’s stuff we should think about–comps give us positive pressure to write that thing we’ve meant to, but it can be negative if we don’t write it very well. But I probably thought of it too much and let that get in the way of developing the NPC stories, which I may do at a post-comp release.

Another is that there is that leap, and I didn’t really sit down and figure more ways to help the player make that leap. And there’s a part of me that has seen relatively newer players figure out the two-bishops mate and they really enjoy having the light go off. (Plus, tutorials are on YouTube now.)

Yet even some masters can find it to be trouble. From my own experience, I read about “some guy rated 2300 who didn’t know it.” It turned out to be someone who had wiped me off the board years earlier, who was born in Germany in 1945 and of Latvian descent. He had written books on chess, which are apparently good. But he had that blind spot!

I really do want to give the experience of “oh man this looks impossible oh wait no I can do this” though and the 2B’s seemed like a good try at that, given how I jumped from not getting it to really getting it. The first part to me felt intuitive, in pinning the enemy king against the board where he can only move two squares, but I always failed at the last.

It seems like the light went on for you, which is what I was hoping to do for the player, but … I probably didn’t help it go on fast enough, for enough people and I missed out on some steps that could/should’ve been added!

5 Likes

Anyone who wants to further improve their two bishop mate skills could try it on Lichess: Piece Checkmates II • lichess.org

(The discussion got me curious, and I’m happy to say I can still do it. Now whether I could do a bishop and knight mate, that’s another matter…)

4 Likes

I was seriously considering this as a post-comp release as a joke, until I realized how brutally hard it could be to not just implement but give clues for and make a story for! Anyway I think chessable’s interface works a LOT better.

B&N course recommendation

Depending on how much you want to learn, I recommend the B&N course on chessable which has a much different approach and final mating pattern from when I was young. It’s surprisingly intuitive and memorable, and it’s the sort of thing that just kind of appears in the computer age. The creator is rated 1900-2000, which is really good, but it’s a reminder titled players can’t figure everything and that magic-seeming nudge to give a stroke of insight/intuition can turn up anywhere.

I’m really impressed by it and how it uses chessable’s interface to funnel/nudge you into the right idea. And it’s free, too!

(Also, only semi-related, another person rated “only” 1900 did an exhaustive analysis of computer notes on queen versus rook endgames. The TLDR is that there are some neat basic guidelines to go by, but the details are hairy. It’s an impressive work. Again, the right interface helped me “see” certain aspects of Q vs R beyond the main winning position when the king’s near the corner.)

3 Likes

Oh, that’s really interesting! I definitely found the hotkey parser idea very interesting and think it’s worth banging on some more. Probably a shorter list of actions would help? Anyway hopefully it came through that I did enjoy my time with the game; as is often the case, I kind of let the review burrow down on one point that I found flawed in an intriguing way, at the expense of emphasizing that it’s a really good time.

Ha, I thought that was a funny coincidence too, especially since I wound up playing the two games back to back (I used a randomizer). I do think the game justifies itself way beyond just the title, though – the theme does make sense, and interleaving the bishops’ lanes of attack does make for gameplay that’s distinct from your previous entries. It’s very interesting to hear that this is a puzzle that doesn’t seem to correlate perfectly with overall chess ability, since it did feel like it was bending my brain in very specific ways.

Oh thank heavens!

That’s a great idea! I put one together of reviews so far, just for you:

The Fortuna
Search for the Lost Ark
Steal 10 Treasures to Win This Game
Xanix-Xixon Resurgence
Xenophobic Opposites Unite

7 Likes

I remember encountering such position in my game. It took me more than 8 minutes to think of the solution. Extremely hard problem for me. Usually, I just promote an extra pawn and be done with the game.

2 Likes

The Last Mountain, by Dee Cooke

Here’s one of my pet theories of IF that I’m not sure I’ve written down before: there should be more parser games about sports. This isn’t due to any native affinity for them – more power to those who are into sports, I could care less about any professional teams, I only did a real sport for two semesters for all my high school and college years, and I’m the kind of schmuck who thinks it’s funny to respond “Interpol investigations” when the check-in question at a work meeting is “what’s your favorite Olympic event?”

No, it’s because of that old writing adage that action reveals character. We can get told that a character is clever or cowardly or chokes under pressure or what not, but until that gets on screen in some way – meaning, in a game, that they take some action that demonstrates the trait – it’s all theoretical. The trouble is, the sorts of character traits that can be revealed by the business of a typical parser game are fairly limited by the medium-dry-goods world model that tends to dominate: “resourceful” and “kleptomaniac” can only take you so far. Then consider that for a linear puzzle game, beyond the difficulty of coming up with and implementing multiple solutions to puzzles, it can also be a challenge for authors to invent reasons why different approaches might actually matter in narrative terms.

Sports offer a fresh way of engaging with these problems: beyond the fact that they create a rules-based framework that supports novel kinds of gameplay, their victory or scoring conditions also offer built-in consequences for a player’s choices, meaning that discrete, relatively-easy-to-implement physical actions can be freighted with narrative and/or thematic weight (This, by the way, is why my dark-horse pick for TV shows that totally should have gotten an RPG is Friday Night Lights). To be clear, I’m not saying that Madden 2023 would clean up at IFComp or anything – but that I do think there’s a lot of potential in parser games that use sports rather than conventional puzzles as their main gameplay elements.

Anyway, I wish that a) I’d written this theory down before playing The Last Mountain, and b) that I could count it as vindication of said theory, when the truth is that it could just be that a talented author like Dee Cooke can make any of their ideas look genius.

Yes, you might have lost track of the fact that this is technically a review somewhere in the previous four paragraphs of maundering, but I swear, these thoughts are relevant to understanding why this Adventuron game works so well, and feels (at least to me) so unique. The setup certainly isn’t one you’ve heard before: the player midway through a long-distance foot race with their running partner, Susan, who’s uncharacteristically flagging early as you tackle the last mountain before the finish line. You’ve got a water bottle, a flashlight, the race directions (there’s an orienteering component), and some walking poles, and with those you need to overcome a series of obstacles – getting tired, losing the trail, facing one last steep descent. Some of them are decision-points, some are inventory puzzles, and none on their own is that innovative – but again, the fact that they’re all happening in a race rationalizes the barriers, and adds a compelling urgency to solve them quickly.

Susan is the other part of the equation. The game deftly sketches your relationship with her – she’s somewhere between a friend and a mentor who helped bring you into this racing hobby – and presents her uncharacteristic fatigue as a central dilemma of the game. Again and again, you’re faced with the option (and Susan’s explicit prompting) to leave her behind so you can get a good finishing time. I’m guessing that most players won’t be tempted to ditch her, but still, the fact that the choice is there lends added weight to the individual puzzles.

The prose thus has to accomplish a lot of different things: create a sense of place, of course, while making sure to foreground Susan’s presence and give the player everything they need to engage with the game. It’s thankfully up to the task, and accomplishes all this with economy and without getting showy, too. Here’s a bit of mid-game scene-setting I especially liked:

As the trees become denser, you realise how dark this forest can start to feel when the daylight isn’t so bright. You’ve never been here so late before. It makes it really difficult to identify the right path, even with Susan’s keen sense of direction.

The forest has become really dense here. The smell of dry branches and the hooting of birds surround you, making you feel a little claustrophobic.

My one kick against the game is that I experienced a few guess-the-verb struggles, partially born of my own lack of experience with this kind of running, but also partially because the parser didn’t feel like it was meeting me halfway as I flailed about trying to figure out how to use the walking poles (yes, for those of you who have played the sailing sequence in Sting, I am aware of the irony of me of all people whining about this). Beyond that, I suppose one might complain that the player will guess what’s up with Susan well before the ending – but I don’t think that’s actually a fault with the game; the tension between the player’s suspicion that it’s serious, and the protagonist’s urge to do well in the race, is another piece of the engine that helps make it work so well. And for all that the reveal wasn’t a surprise, I still felt that it had emotional heft when it landed.

All of this is to say that I found The Last Mountain very good on its own merits, as well as instructive for the directions I think it suggests for future works, which is exactly the sort of thing one wants to come across in ParserComp!

transcription_The_Last_Mountain mr.txt (25.5 KB)

17 Likes

Thanks so much for your lovely review!

I’m really happy that what I intended to get across with this game (which was fairly short in the making but quite a few years in the thinking) has been picked up and appreciated. Running events, especially really long ones, can sometimes turn into such a precarious series of decision-making with so many things that can go wrong, and I spent some time (often while running!) thinking about how those decisions might play out in game form. The Susan character was originally earmarked for another project but ended up fitting this game as a representation of the complicated camaraderie in races.

I might postmortem this one at some point once I’ve done some more reflection. Thanks again for playing and reviewing.

11 Likes