Just pretend it’s a Flash Gordon reference and you’re all good!
yeah i learned on one of those skateboard/scooter combos so there was no way to learn except pushing with my front foot and i refuse to change
I learned to skate from my dad (and I was bad at it so I didn’t stick with it) so I actually never learned to accelerate by pushing with my foot at all. I just did tic-tacs forever
1 Like
I can do an absolutely hideous ollie that always reminds me of the episode where george is describing being called gay by some teens who watched him jump over a puddle on the sidewalk
9 Likes
did anyone play this 1
2 Likes
This is the perfect achievement because it taught me a much more efficient way to play the game. When I played through it the first time, I spent a fair amount of time hand-crafting electric poles, inserters, conveyor belts, etc. When I was shooting for this (never got it but I was definitely on my way, I just never finished that game) I learned that not only is it less annoying to make this stuff in a factory, it’s way more efficient once you set it up. Then it’s just a matter of dropping by whenever I run out of supplies. Now whenever I play, I always do this to some level because it’s silly not to.
It reminds me of what I thought achievements were going to be when I first saw them on the Xbox 360: ways to encourage people to play games differently. Geometry Wars’ “pacifist” achievement was the ur-example of this. It’s a real bummer that achievements are 99% markers of how much one has played a game rather than interesting challenges that change how the game is played.
10 Likes
I’ve been wondering how much developers look to achievements now as a means of collecting cheap post-release metrics. It would give some basic data on how many players did a specific action in the game (although doesn’t fully explain why they did it).
I guess alternate goal/strategy-style achievements would be less useful for this purpose since they only really tell you how many achieved the ‘can they do subversive shit’ metric. I think the teaching through achievements is a good idea but it’s never really built into the game (you usually have to check the achievement list on the platform rather than have it prompted in game). Feels very secret.
5 Likes
This is interesting, I think for small scale developers this might be very true, but I assume that AAA games are collecting way more metrics than what achievements would allow. As a player I think that achievements are currently most interesting as metrics, specifically “what percentage of players got past act 1 of this game?”
Weirdly, I think that Steam has changed achievement percentages recently. Pretty sure it no longer counts folks who have never launched the game. Used to be that whatever the first achievement was 5 minutes into the game would have, like, a 40% achievement rate, and getting the achievement about 2 hours in would be down to 20% or less. Now those numbers seem roughly doubled in most cases.
The most recent Achievement Shock I had was with Dark Souls, where like 30% of people beat that game.
3 Likes
IMO the best use of achievement-like-structures in the current AAA space was Mario Odyssey, except I don’t mean the actual achievements they have. What struck me was the piles of coins on top of hard to reach spaces that served no purpose other than to say “yeah dawg we knew you could do some weird bullshit to get up here.” Winks and Nods from the developer is a nice way of running achievements IMO.
7 Likes
https://nvlabs.github.io/wc-vid2vid/
Its not real time of course but dang this stuff is fast on the way to becoming usable in an interactive context
I like how its at the point where it remembers the general shape and color of an object but forgets many of the smaller details between frames, very dreamlike
8 Likes
I watched this entire video and maybe I’m high but I still don’t understand what this actually is supposed to be doing lol
3 Likes
what i got from it was
- have neural net watch a video
- give neural net first frame of video
- have neural net recreate the rest of the video from that single frame
i don’t think that’s accurate at all but i do find it pleasingly cruel to the neural net
3 Likes
again maybe I’m high but like what possible application could that have lol? is the idea to extrapolate believable video footage from a single frame?
The best extrapolation I can come up with is to build 3D spaces out of videos. That seemed to be a part of it. That has obvious implications for video games but beyond that I was totally lost.
Whatever legitimate purpose this could have, which sounds unlikely, the simplest high value use for this is automatically generated deep fake videos.
2 Likes
Like on a basic level it takes an image with color tagged regions corresponding to object type and generates a photo realistic version.
The important part of this paper in particular is the added use of long term memory in the form of a color point cloud. Unlike previous work it means that a spacial model of the environment builds up in memory and keeps images consistent. They show off how they can do a full loop of the virtual street and end up with the same view as the beginning.
It’s also really easy to generate training data from existing videos, and input data for a model from a simple game environment.
Eventually you can do cool stuff with this like exploring an AI dream world. Or extrapolate full 3d environments from incomplete video data of existing places
4 Likes
so like 40 seconds in they say, imagine using this for realistic rendering for an untextured game, and later they demo it, and explain how they can synthesis left/right eye video from a reference video. they show a synthetic camera shot of a group at a table created from a source shot from a different angle in passing on a slide
really there are three parts to this: the colourised point cloud (from previous output), the flow warping (of the previous frame by the camera movement), and the semantic tagging of image parts (the computer vision part). you’d feed it a low-fidelity flat-shaded render of a scene that looked like the input to that GAN landscape generator, and it spits out a reasonably realistically textured video
the nice thing about this approach is its flexibility: get the semantic tagging from a reference video and generate textures from a different training set and you’d get a consistent style transfer. turn daytime footage into night with rain, remove/add crowds/traffic, realistic false colour for low-light capture. or take the point cloud from a reference video, render a new semantically tagged video from a repositioned camera, and synthesise a shot of the scene from a new angle
4 Likes
hehe neural net go b̷̢̤͖̘̖̈́͋͗r̴̲̖͓̣̩̲̤̹̎̒̈́̐̍͜͝ͅr̸̡͈̘͂̐́̂̀̓͆̒́̓̏́͐̈ṟ̴̥͙̥͎̔̆̋͒̽̀̿̉͜r̸̢̨̛̭̖͍̋̈́́̉̈́̍̏̊̊͝ͅr̸̨̧̧̥̰̮̞͇͈̜̯̫̋̿̀͗̓̉̎̒̀̽̏͌͆̚ͅͅͅr̷̮͖̥͆͋͌̈̓̍̕r̸̛͕͙̝͔͇̹̳̝͍͓͕͋̅̃͆̅̅͋͌͊͑͐̚͝r̷̪̳̙̈̋͒̀͛͑́͘͠r̴͉̀͌̕ȑ̷̹̳͓̤͚͍̺̪̣̜̘͎͖̏̌̀͗̅̔̈̏̅̇͜͝ͅŗ̴̧̗̮̫̜̜̭̱̥̣͂͊̉͝ř̵̮̪̖͍̟̘̖̻̤͚̖̐̅̎̍̀̀̌̇̕ŗ̴̭̣̠̞̄̓̉̕̚͜
12 Likes
They’re developing a machine for creating and sharing new realities made from pieces of our own
We’ve been making slow progress on telepathy for a long time but it’s starting to get weird
9 Likes
Isn’t that just what a novel, or like a roleplaying game is! Less invasive, too.