@wavuwiki siteop, https://xitter.cf/t.co/eTxgtgUA6o and https://xitter.cf/t.co/1TOIfkXm5F creator, @bigcheese_gg tekken TO, dilettante, Sephiroth poaster. Yes, I eat them every day.
Australia
Joined December 2021
- Tweets861
- Following50
- Followers1.5K
- Likes1.8K
If we've analyzed Tekken 8 in depth and have specific critiques, it'd be better to outline them precisely. Things like:
- Hitboxes are computed vectors, not hand-placed 3D colliders, which makes tuning them precisely impossible, c.f. xitter.cf/alcratz/status/2067414…
- Tracking is janky. You'd expect something like "move realigns at X speed on frames Y-Z" but actually there are about 9 different flags and they're things like "homing if on-axis on frame 8", "tracks for first 8 frames", "tracks for 8 frames before first active frame", "tracks during recovery", "inherits tracking from previous cancel", and "completely homing". And these flags can be both on the move itself and on its cancels. Again, this makes precise tuning impossible.
- Pushboxes struggle to keep players separated without also pushing them too far apart during animations and making attacks whiff. Velocities are coupled with animation data, which again makes precise tuning impossible.
- Wall interactions are an overgrown web of hacks and exceptions. Every patch they change how some move behaves at the wall because the underlying system does not generalize well. Related to above issue with pushboxes.
- There is no pure physics free simulation, which makes rollback, spectating, and replays computationally expensive and the UX cumbersome.
Those are major technical flaws that almost everyone would agree on. Then there are things like:
- The large roster with big movelists creates an onerous knowledge burden. On its own, this is fine, and even good, since it gives the game depth. But then Tekken more than any other FG has a duty to proactively teach the player, and yet:
-- The story mode doesn't teach the player how to play, c.f. cthor.me/Story-mode.
-- There are no cues that e.g. distinguish highs from mids from lows, safe moves from unsafe ones, strings with jailing or force block highs from duckable ones.
-- There's nothing teaching players how tracking or hitboxes work, because as previously mentioned... they kinda don't.
-- The built-in training routines are limited and (aside from the throw break practice) aren't serious exercise tools. This isn't something you can assign one developer to for a month and call it a day. Takes a lot of craft to make it even mildly useful, c.f. xitter.cf/6weetbix/status/209605… for a bare minimum prototype for one singular lesson.
- The FT2 ranked system discourages playing interactively and encourages running gimmicks.
- The Tekken Lounge is public and unmoderated. You can't make a private instance to e.g. organize a clan, social group, or tournaments without getting gatecrashed.
- The "time to fireball" – how long it takes to open the program and throw an attack – is one of the worst in the genre. Makes it annoying to lab.
Those are some big product issues. Then, matters of taste regarding the actual battle system:
- Too much time where players have no agency. Freeze frames, cutscenes, overly long combos, and jailing strings with no counterplay (many heat moves) all contribute.
- Relatedly, too many situations lack real-time pressure, making the game feel turn based.
- Too much easily looped offense, e.g. Steve LNH, Raven BT.d+2, Kuni FC.1+2, Law and Shaheen slides. More generally, too many ultra plus lows.
- Many stance transitions don't have layered counterplay. Sequences are too repetitive.
- Not enough neutral resets. Once pressure has started, it rarely ends. It just switches polarity.
- Neutral is warped around demon paws, running moves, and safe launchers. Other move categories get powerkept to keep up. Now everyone's neutral is rotating god buttons without regard for the opponent's patterns.
- Heat Burst is strange. It presents as "sacrifice meter to enter heat under pressure", but in practice it's mostly a combo tool. Outside of combos, it's riskier than a regular powercrush, because the freeze frame makes optimal punishment easy.
- Heat Dash on block can't be tuned. The "always +5" rule means almost every character relies on a single heat dash button: whatever one is most unsafe normally.
- Execution barriers are poorly tuned. Most are either too easily mastered or not satisfying. Nobody cares about all these "blue spark" moves for 1-2 extra damage.
- Many i10 punishers are too rewarding. Not enough reward for punishing precisely, and not enough of a risk gap between -10 and -14.
- Risk gap between -9 vs -10 and not-launch vs launch is enormous, making risk tuning imprecise.
- Main way to use recoverable health for a comeback is just "hit a launcher". Other ways to recover health are insubstantial or unreliable.
- Recoverable health makes it rare for both players to be low health: the conditions for slow-mo, GREAT, and highest tension situation in any FG.
I could bandy this list and go around proclaiming "Aha, look, Tekken 8 is bad game!" But if I knew any other game as well as I know Tekken, I'm sure I could make a list just as critical for that too. And if I had the time, I could make a list of all the things that Tekken 8 does well that's 100x longer.
We shouldn't reduce our analysis to "Tekken 8 is low quality" because that doesn't let others weigh up or dispute each of these points. One could very well look at this list, agree with every point, and still think the game is good despite them, because the good outweighs the bad.
Many things in life are like this. What you choose to focus on often holds more power than any underlying truth about the thing being examined. Focus only on the negatives, and you will see a bad game.
YouTube viewers are not such ruthless critics, so the same doesn't really apply there. And for any given video 99.99% of potential viewers have never heard of it, so marketing would move the needle massively. Many indie FGs are still in this position, but a game that was in the main EVO lineup is not.
But still, yes, it'd be foolish to think that e.g. MrBeast does not demonstrate immense craft in his work. Anyone looking to make YouTube videos should study his work with curiosity rather than sneering. That doesn't mean his work is tasteful, but a tasteful Youtuber still does ~95% of the same things. So it'd be more incisive to say "MrBeast/T8/SF6/Strive are crass" rather than calling them bad games, if we want to be curt.
Probably the reason we don't do this is that popular things are usually crass. That's how taste works. It doesn't take a keen eye to notice MrBeast is crass. So we are tempted to inflate our take to something like "fundamentally flawed" or "broken" or "doesn't function" or "low quality" or just plain "bad", but this is both overplaying the hand and I think dismissive to people who still enjoy the game, which is a lot more people than Twitter discourse would indicate.
xitter.cf/alcratz/status/2101373…
Fighting game players are ruthless critics. If a game has even the slightest flaw, you won't just hear about it endlessly; it'll show in retention figures. Devs need the sun and moon to align just to retain 5% of their players 1 month after release.
It's good we have these standards. It forces devs to focus on gameplay first. No studio would dream of making a FG today that wasn't first and foremost aiming to have a good battle system. Most fail because the standards are that high, but they're still trying. The same can't be said for other genres.
That's the flattering part. Now for the reality check. If this is remotely true, you have to accept that Strive, Tekken 8, and Street Fighter 6 are, as a whole, good games. Not just good products. You might have major issues with certain aspects of them; I certainly do. They might not be to your taste. But they are bona fide fighting games with retention figures impressive not just within the genre but the entire industry. If they really were as shallow as their critics imply, they would not still be this big 3+ years after release. Players would be bored by now. Fixating on the same 2-3 issues elides the hundreds of other things they get right.
And an even harsher reality check: whatever niche fighting game you want to go to bat for probably isn't that special. It didn't distinguish itself well enough. I won't name names, but if your game ever appeared at EVO and struggles to crack 100 online players and has no locals, it isn't some hidden gem. It was weighed and measured and found wanting. No amount of "content creation" or "community leadership" (i.e. marketing) can save them. It's dead. Let it go.
The silver lining is those dead games are probably pretty damn fun still. Not surviving the cutthroat FGC gauntlet isn't a scarlet letter. You've only gotta find one other person who loves it as much as you to play. If you really love the game, you can make it happen. Most competitive genres are not like this! There's a mass grave of competitive games that you really can't play because filling a whole lobby is a herculean feat. (That's before getting into many games being literally unplayable from the live-service going down.) Your locals can't save the game, but they don't have to. Just play.
Carlsen doesn't win by predicting your moves. He just plays the least bad moves, and you make worse moves. His strategy is simple: he won't hang pieces, will see more and deeper tactics, and has more positions memorized.
But a stacked board is a fair challenge to the analogy. An adversarial ASI starts disembodied, shackled, and with few allies, so you could grant that misaligned ASI is possible and still think humanity is safe. Stockfish can never win with just a king, even against Martin.
Chess was conquered early by AI because all you need is a good opening/endgame database, a decent search heuristic and eval function, and some compute. You don't need to model the enemy at all, just the game state. Go fell the same way. AlphaGo was just better search, more compute, and a self-learnt eval function. But these are two-player games with no one else to win material from.
So let's move the analogy to poker. What happens if we stick one GTO bot, one elite human, and six noobs at a no-limit table? The elite player wins most often. He exploits the noobs better, gets a bigger stack, and then wins the one-on-one off chip share. The GTO bot still wins pretty often, but even the noobs can win occasionally if they get lucky. To win more at this table, the bot needs to exploit the noobs harder. It needs to model their behaviour and find adaptations that maximize winnings while minimizing leaks.
Let's say an ASI does just that, and we are the elite player. Not only do we now have to beat a souped-up GTO bot one-on-one, but we have to exploit the noobs harder as well. In doing so, we may end up outside our comfortable range and get exploited too before even getting to the much simpler one-on-one.
Still, perhaps we get lucky and end up with a bigger stack for the one-on-one. The ASI has a consistent edge. If it could control the stakes, it could guarantee a win: always hold some chips in reserve. As long as it can keep playing hands, eventually the edge overcomes our chip stack, no matter how big it is.
No-limit poker doesn't quite allow that, though. The big stack has a structural advantage: shoving. The short stack doesn't get to control the stake while keeping its edge. As soon as it puts chips down, the big stack can force an all-in. Keeping the edge requires calling sometimes, but that risks everything on a single hand.
For us, trying to close the edge by playing slow and learning the ASI's leaks is losing, since doing so decreases variance. Worse still, the longer the game goes, the more the ASI learns about us. And since it's smarter, the edge only grows the more time it gets. Our only sensible strategy is to force the issue. This is "just unplug it": shove aggressively and early, and we win almost every time off the massive chip lead.
Now, this is granting both that ASI is possible and that alignment isn't. But assuming this, "just unplug it" and relying on starting conditions is an awful plan.
Let's say you enter the table with a mountain of chips. The noobs get stacks of various sizes, and the ASI gets a short stack. It folds whenever you enter the pot, so you never get to shove. Its plan is to farm the noobs first, then fight you on an even footing. How many noobs do you think are on the table?
Yudkowsky’s Magnus Carlsen analogy for AI takeover serves his conclusion extremely poorly. If anything, it makes me believe him less.
The thought goes like this: “Carlsen can see at least what you can see on the chessboard.” Therefore he will anticipate your plans and stop them. This is then used as an analogy for why you cannot stop a superhuman AI by denying it resources. After stymying your plan, he/it will beat you with a winning move you do not see.
Look closely and the analogy fails to prove what he wants it to prove. Worse, it fails in a way that suggests how his thesis could be wrong in real life. It is anti-persuasive.
I am not sure how many times he has used this analogy, but it is very popular among his followers, and he has used it here about three times in the last week.
The problem with evoking Carlsen and chess is that it only works in a particular range of positions. It implicitly assumes a board equal enough (or favorable to Carlsen) that his better play wins out.
In chess there are plenty of positions unrecoverable for Carlsen. He has been in thousands of them in his life, and you could devise any number more in which he would resign on the spot.
In fact, I'll do you one better: we *already have* AI superintelligence in chess, and it also finds many positions unrecoverable.
A superintelligent chess engine really *does* anticipate your moves perfectly. Given a bad enough position, it knows for sure, say, that your rooks will roll down like a curtain, cutting off space to its king, until the king is driven to the edge and the last rook takes the last rank for checkmate. It will evaluate such a position as, say, “M6”: White will win in six moves and there is nothing to be done. Being able to foresee a plan does not entail being able to stop it, in chess or in life.
Would a superhuman chess AI play its way into an unrecoverable position from the regular chess start position? No; it's too strong and reliable. But given a position already in play, Stockfish will lose in many positions where intelligence is simply not the dominant factor.
In life, any “rogue” AI inherits an already-in-play board—one in which the material conditions (property rights, command of political structures, and so on) disfavor it substantially. Chess actually shows that intelligence alone does *not* make a position winnable, the precise inverse of how Yudkowsky invokes it.
Fantastic GDC talk by former 2XKO Champ Lead @Blinkity gdcvault.com/play/1034812
Understand a genre's mythology and a lot of odd conventions suddenly make sense. You stop breaking things that work and start actually innovating.
If Riot deserves any blame at all (and that's a big if), it's for not forcing 2XKO out the door sooner. Game dev isn't a charity. Creatives need to be herded a little. You gotta make something the market actually wants, and your budget needs to match your TAM.
Who bronzeman mode is for, research into 2009-era proto-ironmen and how their attitude differs, how a 1978 philosophy book ties it all together, and a revised bronzeman that unites players under one banner instead of dividing them further. cthor.me/Tinman
What is skilling's identity? To answer this, I've spent the last two months listening to skillers. The result of my analysis: four Jungian archetypes, a map that can guide content design, and an explanation for why the gnome cube debate never ends: cthor.me/Pillars
