乐尔乐

[乐尔乐 @leerle777]

Recommended Working Pipeline

[SpaceX @SpaceX]

Watch Starship Flight 14 https://t.co/uQKQvgbrch

[How To Prompt @HowToPrompt__]

NVIDIA built a AI that nukes 30 years of animation technology.

They open-sourced a model that generates 350,000 motion skills in real-time, at 15,000 fps with just 2ms latency.

It’s called MotionBricks, an AI framework built to generate real-time character and humanoid motion at scale.

→ Intuitive smart primitives for quick scene authoring.
→ Integrated into NVIDIA's GR00T robotics stack.
→ Zero mocap. Zero rigging.
→ Runs in real-time

[乐尔乐 @leerle777]

From Tool Use to Tool Creation

Going further, AI shouldn't just be able to:

Use existing tools.

A truly advanced agent should be able to judge:

That a suitable tool for solving the problem is currently missing.

[乐尔乐 @leerle777]

testcode 第一次长任务顺利运行 🩵

[乐尔乐 @leerle777]

Heads up for Codex CLI users: avoid 0.143.0 for now.

Related issue: openai/codex#31665

With GPT-5.5, Codex may fail before shell execution:

unsupported call: exec_commandexec_command

Even pwd / ls can fail before reaching the shell.

Workaround:
codex -m gpt-5.4

[乐尔乐 @leerle777]

@NVIDIAAI As shown, what caused the image quality score to actually decrease?

[乐尔乐 @leerle777]

@shulmnn @hookuru_ Yeah, I noticed that too. The immersion feels much stronger now. A quick bullet-time beat before the final cut-in—hazy at first, then snapping back into full color—could make it feel like coming back to reality.

[乐尔乐 @leerle777]

@Vivek4real_ @grok How is this curved line generated, and how did you create such an elegant curved road editing tool? How are the boundaries stored? Does it support dynamic snapping and matching to the curved road edges, or is it done by manually dragging?

[乐尔乐 @leerle777]

@RomanF65805 @hookuru_ Exactly

[乐尔乐 @leerle777]

This shift represents a move toward higher-dimensional AI experiences, where models are beginning to genuinely understand and navigate 3D environments.

[乐尔乐 @leerle777]

These days, AI is reaching a new stage. Models such as Fable 5 or GPT-5.5 and other frontier systems are approaching the level of intelligence needed for many tasks performed by the general public.

[乐尔乐 @leerle777]

Some of these worlds may run inside game engines. Others may be rendered directly in the browser with technologies like WebGPU, Three.js, MCP-connected tools, and similar platforms.

[乐尔乐 @leerle777]

Stop being only a content consumer.
Less passive watching and reading.
More commenting, sharing opinions, and turning thoughts into meaningful output.

[乐尔乐 @leerle777]

Starting today, I'll try to post more in public.
I want to share my AI projects, product ideas, and learning process with more people.
Build in public. Learn in public.

[乐尔乐 @leerle777]

lv1-lv3 I have a triple-layer safeguard to ensure my tokens run smoothly. For tiers LV2 and LV3, I use a custom-built, LiteLLM-style proxy to handle the varying formats across different models—since standard token limits fall short and the Pro plan is too expensive.