LLMs trained using a universal language

<!-- SC_OFF --><div class="md"><p>What if there was a new programming language where the meaning of each token was so dense (or perhaps so specific) that an LLM could write robust

<!-- SC_OFF --><div class="md"><p>Other than for safety reasons, why haven't AI models that think in their own optimized &quot;alien&quot; language been developed before? Wouldn't

<!-- SC_OFF --><div class="md"><p>I think LLMs have trouble with conceptual generalization because every dimension of their representations ALWAYS participates in every operation,

<!-- SC_OFF --><div class="md"><p>Every &quot;AI agent&quot; that controls a computer today does this: generate text → parse it → execute → serialize result back to text → repeat.

<!-- SC_OFF --><div class="md"><p>I've been learning German recently, and it occurred to me that I could point some of my AI horsepower at having a German speaking LLM to practice

<!-- SC_OFF --><div class="md"><p>I've been using STT models and noticed there are specific models for things like English. I've wondered why we haven't had the equivalent for

<!-- SC_OFF --><div class="md"><p>If LLMs are gonna write most of our code anyway, why are we still making them write Python/Java/Rust?</p> <p>Why not make a language specifically

<!-- SC_OFF --><div class="md"><p>Hi everyone,</p> <p>I’ve been using ChatGPT’s voice mode quite frequently, and it’s incredibly effective, especially for conversations and