AI Language Optimization and Efficiency
On 21 Jun, singularity asked why LLMs could not be trained to think in an optimized AI language that replaces English as their internal medium. The question framed human language as a possible efficiency constraint on machine reasoning.
By 11 Aug, singularity had extended the idea to software development, asking whether AI could create a coding language designed for extreme token efficiency. The proposal remains conceptual. The supplied signals contain no working grammar, language specification, implementation, benchmark, or research result.
Other industry discussions address the same premise. MachineLearning asked whether a dedicated programming language for LLMs would be viable. LocalLLaMA raised the prospect of LLMs trained with a universal language, questioned why AI controls machines through human text, and pointed to neural networks writing their own update rules. OpenAI carried Harry Collins’s argument that LLMs cannot do frontier science because they cannot invent new language. ArtificialInteligence asked whether intelligence needs a body and why AI has not derived answers to the universe when mathematics is its language. Singularity separately asked whether AI could fully solve the language barrier.
Assuming the claims are factual, impact is 0%. Nothing changes for LLM developers, programmers, or users because the reported activity consists of speculative questions. There is no tool, model, standard, or demonstrated efficiency gain to adopt.
Watch for a published language specification, a working model trained to use it, or benchmarks showing lower token use while maintaining performance. Those would turn the discussion into a testable engineering claim.
<!-- 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>I’ve been thinking about the current AGI discussions, especially the debate between scaling language models and building embodied systems.</p>
<!-- 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>Every "AI agent" that controls a computer today does this: generate text → parse it → execute → serialize result back to text → repeat.
<!-- SC_OFF --><div class="md"><p>Are there any solutions where a large language model isn’t trained on data in specific languages—such as English, Spanish, French, etc.—but is
<!-- SC_OFF --><div class="md"><p>I’m sceptical that AI will ever completely remove it, mainly due to latency and other constraints, but also because language is not just
<table> <tr><td> <a href="https://www.reddit.com/r/LocalLLaMA/comments/1sez79x/towards_a_bitter_lesson_of_optimization_when/"> <img alt="Towards a Bitter Lesson of Optimization:
<!-- SC_OFF --><div class="md"><p>I think this is actually what AI provides. A way to give a computer instructions using human language (no coding knowledge needed). </p>
NewsHive monitors these sources continuously. All signal titles above link to the original reporting.
Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →