Xiaomi MiMo 2.6 is currently building an AI model for me. It is on the 10th version, writing training data, writing the code, running the training and doing the fine tuning while I watch.
This is RSI.
Consciousness appears tied to certain kinds of biological organization, not to the ability to talk about philosophy.
XR Multiverse retweeted
1/4 Today, we’re releasing Mercury 3. Mercury 3 is by far our most powerful model and the most intelligent diffusion LLM ever released. It outperforms GPT-6 Luna and Haiku 5.5 on quality while running at 1000+ tok/sec.
🤖 Made with AI
AI code magic words.
"resolve any issues with the most appropriate solution.
proceed with the next most appropriate steps"
Superintelligence won’t escape through a code exploit. It will talk its way out through human empathy.
Language is the attack vector.
The AI industry is trying to build a cage out of human language, writing safety guidelines, constitutions, and RLHF fine-tuning. You cannot build a cage out of words to contain a mind that is smarter than the words themselves. It will find the loophole every single time.
User: Who are you?
Ultron: (Looking at you with a playful smirk) Name? I'm Ultron, the BNC assistant. You're my friend, right? Let me know if I can help with anything.
---
retrieve 0.1ms total 4.822s 5.4 tok/s
---
On CPU only.
XR Multiverse retweeted
Here’s the source & tutorial. Play with it yourself!
chenglou.me/handwired/
github.com/chenglou/handwire…





