liang rao

discovering, explorering

[liang rao @cat3dog4]

@dr_logvinovich @grok 发生了什么？

[liang rao @cat3dog4]

@Mileson07 @nangkengrou 写死判断具体怎么理解？还是我前面那个渲染代码的例子？需要怎么写死？刚申请waitlist还没拿到测试权限，哈哈，只能先咨询一下。

[liang rao @cat3dog4]

@Mileson07 这个场景可以吗？：一段完整渲染代码，让它判断有没有错误，能不能正常渲染所有公式，效果，交互。有错误返回给LLM继续修改，能跑通则直接渲染。

[liang rao @cat3dog4]

Runnable in the browser. Reusable from npm + PyPI. Dataset on Hugging Face.

The wiring is connectome-derived; perception + neural dynamics are explicitly modeled.

Try it → https://t.co/bnP2tak2Mj Code → https://t.co/uggZAU4Raj

[liang rao @cat3dog4]

A fly brain can now see your world.
Demo: https://t.co/yBGN04vDAD

I just released FlyEye: an open developer kit + browser experiment built on real Drosophila connectome-derived wiring.

3,376 neurons. 78,797 signed edges.

#connectomics #neuroscience #opensource #webdev

[liang rao @cat3dog4]

This inspired me to build my own version of the game. Really love the idea — thank you @yungcontent for the inspiration! 🙌

I’m taking the same core concept and exploring it in my own way. Excited to see where this goes. 🎮🤖

https://t.co/h767FoBXQ5
https://t.co/mWli7q9JMA

[liang rao @cat3dog4]

I’m asking whether we need an invariant Human Approval Boundary — one that the recursive process itself can never modify or bypass.

That question may matter more than when AGI arrives.

[liang rao @cat3dog4]

Jacob Coxon’s @hilbertspaess warning about recursive self-improvement raises a question I can’t get past:

Where is the human approval boundary?

The issue isn’t whether AI should be allowed to write or modify code. AI already does that.

[liang rao @cat3dog4]

If we wouldn’t let an AI autonomously rewrite critical production systems without safeguards and approval, why would we accept weaker oversight when it starts rewriting the system that creates the next AI?

I’m not against self-improving AI.

[liang rao @cat3dog4]

But with recursive self-improvement:
AI → improves AI → stronger AI → improves the next AI → ...

At what point does meaningful human review disappear from the loop?

[liang rao @cat3dog4]

The issue is whether AI should be allowed to modify the very mechanisms that determine how the next generation of AI is trained, evaluated and improved.

AI writes code → humans review it. AI improves an AI → humans evaluate it.

[liang rao @cat3dog4]

I put humans and 6 top LLMs (Claude, GPT, Gemini, DeepSeek...) into the exact same 3D Shanghai promenade.

The catch? You walk up to strangers, chat for 5 rounds, and have to guess: Human or AI?

https://t.co/1APz0rLfWO

[liang rao @cat3dog4]

Humans are trying their best to act like bots. Bots are pretending to be tired humans who typo and say "gtg".

Can you tell who is who? 🎮 Play directly in browser (Mobile & PC supported): https://t.co/1APz0rLfWO

#AI #TuringTest #IndieGame #Web3D #Threejs #ChatGPT

[liang rao @cat3dog4]

@Atharv_Golait https://t.co/oRDRe97n57 ill send you more info privately.

[liang rao @cat3dog4]

Looking for an Experienced Web Application Security / Penetration Testing Specialist — $15,000 Reward

We are looking for an experienced security specialist to conduct an authorized security assessment of a designated web platform.

[liang rao @cat3dog4]

Required expertise: Web penetration testing, API security, authentication/authorization vulnerabilities, OWASP Top 10, and professional vulnerability reporting.

Testing is strictly limited to the explicitly authorized target and agreed scope.

[liang rao @cat3dog4]

Compensation: USD 15,000, subject to successful completion of the agreed assessment and deliverables.

We expect clear proof of findings, vulnerability analysis, impact assessment, and remediation recommendations.

[liang rao @cat3dog4]

Objective: Determine whether security vulnerabilities could allow unauthorized access to:

Backend user accounts or account data
User platform/account information
User balance, credit, top-up, or transaction-related information