KORVA (@Jocker_zero) header
KORVA (@Jocker_zero) profile picture

A gamer, then an engineer Game Dev · AI · Tech Indie game developer 前网易、字节工程师

People's Republic of China
Joined May 2016
Photo by KORVA (@Jocker_zero)
An Indian woman records her husband watching other women more than 50 times. The man looks so desperate 🇮🇳. Game-industry context: this reaction loop is also familiar in game design. Repeated exposure can turn a simple stimulus into a compulsion, so systems that reward attention need clear limits and readable feedback. The challenge is to create that pull without treating the player as a behavior to extract. #GameDesign
2
2
107
Docker Desktop has long meant lightweight containers running inside a multi-gigabyte VM, with noisy fans and a separate enterprise subscription. Apple has now open-sourced `apple/container`, a Swift tool under Apache 2.0 that takes a different route on macOS. Instead of placing every container inside one large Linux VM, it uses macOS 26 native virtualization for one micro-VM per container. A minimal vminitd process inside a stripped-down kernel communicates over vsock/gRPC, aiming for hardware-level isolation without losing millisecond cold starts. Rosetta 2 handles amd64 images. The official container-examples repository includes containerization-sandboxy, showing how to run an AI coding agent such as Claude Code inside a micro-VM with only selected folders mounted and network access limited by an allowlist. For game teams, this is more than another Docker alternative. AI agents increasingly touch build scripts, asset pipelines, internal tools, and local project files, so a disposable VM boundary creates a clearer blast radius when something goes wrong. The project currently shows 51.5k GitHub stars. If you work on a Mac, are you staying with Docker Desktop, using OrbStack, or moving to Apple's native container? #AppleContainer #GameDev #Docker
3
1
2
264
The newest X ranking notes include a constraint game developers should design around: For You candidates older than 48 hours are filtered, while a bounded cold-start lift can apply to young posts from smaller accounts. That does not guarantee reach, but it does make delayed publishing and repeated near-duplicates poor bets. A launch trailer posted once and left to age is not a distribution plan. Break the work into timely evidence: a specific bug fixed, a before-and-after benchmark, a playtest question with the current build. Publish while the result is fresh, and vary the format so the feed does not receive the same asset with a new caption. Which result from this week is concrete enough to publish before another day passes? github.com/xai-org/x-algorit… #GameDev #GameMarketing #XAlgorithm
1
72
This is nature at its most dangerous and terrifying. When a helicopter flies over the middle of a river in the dense Amazon jungle and this scene appears below, even the pilot is shaken. That is the feeling I look for in giant sea-monster fights: water hiding the true shape of the threat, a scale that makes your weapon feel tiny, and the realization that the environment was never safe. What is it like for you to fight a giant sea monster in a game? #GameDesign #SeaMonster #Amazonia
3
6
419
A live game has a 90 MB balance update, one million daily players, and a remote-config switch that can change reward rates without a client patch. You can ship the new economy to 1% of players for six hours, 10% for the next day, then everyone. Would you gate the rollout by player ID hash, account age, or progression tier first? I would start with a stable hash so the same cohort stays comparable, then inspect median session length, item sink rate, and support tickets before widening. A fast rollback changes the risk, but only if the cohort boundary is stable enough to diagnose. #GameDev #LiveOps #Production
1
4
299
AI-assisted game production gets risky when generated code becomes the foundation simply because it compiles first. A three-month ship cycle can hide the real cost: the quickest system is often the one you understand least when content, edge cases, and live fixes arrive together. The practical boundary is to make rules, data, and acceptance checks visible early. Keep weapons and enemies data-driven, isolate generated systems behind small tests, and require every agent task to end with a replayable failure case. Speed is useful only when the team can still explain and repair the result. If an AI-generated subsystem passes its tests but nobody on the team can trace its state transitions, do you ship it or schedule the refactor now? #GameDev #AIAgents #Godot
4
2
236
Identity Confirmed From a game-industry perspective, this is a useful reminder of how much a reveal can do with very little text. The clip carries the moment, while the caption leaves room for the community to react and speculate. That can build attention, but the eventual payoff still has to justify the buildup. #GameMarketing
1
89
有人不理解为什么GTAVI的热度这么高,看看下面这个游戏就能理解了 这是R星的上一款游戏,七年前的游戏 #GTA6 #GTAVI
1
2
128
最抓人的不是“AI 做游戏”,而是那个有点离谱的反差。 一款 2D 像素小游戏,Steam 上线 48 小时卖了 5 万份,收入 35.6 万美元,最高同时在线 9,377 人。 我本以为背后至少是个成熟团队。结果帖子里的转折是:开发者几乎是编程新手,Demo 阶段大量借助 AI 完成。 可真正让我停下来的,不是这些数字。 游戏的核心居然只是一个 3×3 老虎机:转出角色,合成升级,自动战斗,再来一局。 没有 3D 开放世界,没有电影级角色,也没有复杂物理系统。 它抓住玩家的东西很朴素:下一次会不会抽到更好的?这套阵容还能不能再撑一关? 看到这里我才觉得,AI 给独立开发者最大的价值,可能不是把画面做得更炫,而是把“这个玩法到底好不好玩”从几个月缩短到几天。 玩家不会因为模型更强就掏钱。他们只会因为下一局看起来更值得,再点一次开始。 所以真正值得研究的,也许不是 AI 能不能做出一个世界,而是它能不能帮你更快找到一个让人舍不得关掉的循环。 #Slotbound #独立游戏 #游戏设计
6
3
349
A gamer with nothing else to do used hundreds of armored vehicles to try to stop a train in GTA V! #GTAV
1
3
472
有点像合金弹头的S枪,手感很特别
1
52
看到这个海浪,来感觉了 让我想起以前实现过的基于快速傅立叶变换的海洋效果 以后有机会要做一款跟海洋有关的游戏 #FourierTransform #大海
87
每次看到这种丧尸追逐逃生场景,都感觉很有带入感,就好像是我自己在跑
1
40
这事情我也干过,但是我只有4辆车 4个大学舍友,每人开一台卡车过来 #GTAV #Gaming #OpenWorld
38
GameLens points at a better AI game-dev pattern than simply giving the agent more context. Its developer reports 32x fewer characters exposed than the native MCP across seven real questions, then pairs that narrow view with an F9 bug key and visual checks. The attached examples flag "nothing behind 3.6% of the frame" and a floating object as "NOT RESTING." That is the boundary I would copy: keep the agent's read surface small, make failure capture one key away, and let a visual regression gate reject a wrong fix. The validation loop matters more than the chat surface. Would you capture the editor state before or after asking the agent to propose a fix? teapaw.org/cosmintrica/status/212… #GameDev #AIAgents #UnrealEngine
Now that the jam is over, I'm fixing what held me back. In the jam my tests passed while the game was wrong: numbers lie when nobody looks. So I started building GameLens: Claude's eyes, keys and memory for Unreal Engine 5.9. One night in: 32x fewer characters to read than the native MCP on 7 real questions, a bug key on F9, and a DLSS test that needed five fixes to get right. Pictures below, details in the replies. 🧵 Thanks to the jam's hosts and partners, and to everyone I talked with along the way: @UnrealEngine @EpicGames @FNCreate @itchio @TimSweeneyEpic @claudeai @AnthropicAI @Google @OpenAI @SpaceXSI @cursor_ai @AMD @diversion_hq @fal @DeemosTech @intel @MeshyAI @NetApp @nvidia @NVIDIADeveloper @NVIDIAGeForce @reallusion @hellorokoko @tripoai @awscloud @googlecloud @elevenlabs @OddberryStudio @dav01da @MrCollison @mattjohnston_ai @chrisfirst @nazimadakli @DavidRaymondRP @SimonsFury @mikerconnelly @nikmlnkr @PashaIGooD @HollyValzNorth #MegaJam #EpicMegaJam #UE5 #gamedev #indiedev
1
170