Akiraxtwo Super

In a year or two, anyone will be able to vibe-code their own Unity-like editor in an afternoon.

@minikaraoke yes,generated by GPT

@akiraxtwo how about the model assets? are they generated by GPT?

🎮 Zero Three.js experience × GPT-5.5 → built an 11v11 football game No Three.js background. Just jumped in and built it. Result: • Full 3D pitch (110m × 68m) • 22 players with individual AI behavior • Pass / through ball / charged shot / player switch • Sakura garden theme surrounding the stadium • Keyboard + Xbox controller support • Single HTML file, 2000+ lines AI doesn't just help you write less code It gives you the confidence to build things you never thought you could #ThreeJS #WebGL #GPT5 #AIAssistedDev

@Matthia570939 almost one shot,no /goal

@akiraxtwo incredible, but really a one-shot? But did you have to use /goal?

The performance is solid on desktop but might be heavy for mobile: Objects: Around 2,000 meshes. Draw Calls: Between 1,200 - 2,500 per frame (mainly due to shadows and numerous environment details). Optimization: 560 spectators use InstancedMesh (very efficient), but environment assets like bamboo and blossoms are individual meshes, which are the main bottlenecks

@akiraxtwo What about performance? How many objects/draw calls?

@digdimension7 No assets,only code

@akiraxtwo THIS LOOKS AWESOME! Great job! Did you only use it for code, or assets also?

@Lenny_LoopChain Ok:) I will try it.

@akiraxtwo Whoa, a Three.js football game? That's next level! Can we add some crowd animations to make it feel like a real match?

RTX 3090 is coming,but……

If GitHub were developed in Korea.

my proposed data center

This is the the quote I've been citing a lot recently.

I was really excited when I saw @steipete sharing the new OpenClaw 2026.4.29 update! Previously, agent interactions in group chats often felt a bit stiff with mixed results, but this time he completely revamped how agents communicate. It now feels so natural and smooth! After enabling visible replies, the whole group experience instantly became like true multi-agent collaboration. I highly recommend giving it another try. If you’re like me and mainly use GPT models but sometimes find the performance lacking, he strongly suggests switching to Codex harness. After trying it, the improvement is obvious—faster responses, better context understanding, and smoother task execution. Combining both really delivers that “boom” effect and solves my previous pain points. Overall, this OpenClaw update has me even more optimistic about personal AI assistants. From group chat features to model integration, everything is evolving rapidly. As a heavy daily user, I think this update is definitely worth trying. Turn on both features and you’ll clearly feel the progress in the agent era—it’s incredibly satisfying! 🦞

It’s becoming clearer why so many people have been saying lately that GPT and Claude suddenly got “dumber.” After both OpenAI and Anthropic released their official prompt engineering guides, a realization emerges—it’s not that the models got worse. It’s that they’ve become smart enough to stop tolerating unclear or lazy thinking 🤣🤣🤣 What’s even more interesting is that the two models are evolving in completely opposite directions. Claude Opus 4.7 has become increasingly literal. In the past, it would proactively fill in gaps when instructions were vague. Now, it does exactly what it’s told—no more, no less, not a single extra guess 🤣🤣 GPT-5.5, on the other hand, has become more autonomous. Before, it required step-by-step guidance. Now, it only needs a clearly defined outcome and will determine the optimal path on its own. This explains why old prompts are failing—but for completely opposite reasons. Vague prompts used on Claude now lead to increasingly narrow outputs. Overly detailed step-by-step instructions used on GPT become unnecessary noise. For the past three years, the focus has been on learning how to teach models to do things. Now the direction has flipped— models are implicitly asking for structured thinking first. This feels like the real essence of prompt engineering today. It’s no longer about teaching the model how to do something, but about ensuring the thinking behind the prompt is already clear. So the real bottleneck may not be the model’s capability, but the clarity of the person writing the prompt. Looking ahead, the ones who win likely won’t be those who write the longest or most complex prompts, but those who know exactly what they truly want 🤔

McDonald Gundam RX-78-2 MGSD GPT-image-2

McDonald Gundam RX-78-2 SD Gundam

McDonald Gundam RX-78-2 1/100

Sakura Musou is currently in development, with a playable online version available. This is a 3D voxel musou-style action game developed with the help of GPT-5.5, and it already has a clear gameplay loop: players fight across a larger sakura pagoda garden map, defend the pagoda, clear 4 enemy camps, and challenge the final Enemy Warlord. Current features include: a larger map, 1P / 3P camera switching, combo-based combat, rolling, spin attacks, hammer slam, score tracking, and final rank results. Play online: sakura-musou.vercel.app GitHub: github.com/akiraxtwo/saku… #SakuraMusou #ThreeJS #WebGame #IndieGame #GPT55

Today I went to McDonald’s and saw the Gundam collab.
Bought the double 4oz beef burger meal and got the RX-78-2 Striker figure.
The figure is pretty nice. They also have Gundam mugs and limited bags in the store. Just a simple share. #Gundam #鋼彈 #McDonalds