LLMs may not need human-style language.
Rohan Paul Twitter · Rohan Paul (@rohanpaul_ai) · 2026-06-25
An arXiv paper introduces BabelTele, a compressed cross-lingual notation for LLM-to-LLM communication that retains 99.5% semantic fidelity while reducing text to 27.9% of its original length, arguing that human readability and machine comprehension are separable properties.
Extraction
Topics: token-compressionllm-efficiencymulti-agent-communicationllm-research
Claims
- BabelTele achieves approximately 99.5% semantic fidelity while compressing text to 27.9% of its original length.
- Human readability, natural-language fluency, and machine recoverability are separable properties in LLM communication.
- Future AI systems could save context space by using dense model-readable messages instead of natural-language prose.
- Human prose carries redundancy that models trained on large symbolic mixtures may not require.
Key quotes
In the paper's strongest result, BabelTele keeps about 99.5% semantic fidelity while shrinking text to 27.9% of its original length.
Human prose carries redundancy because humans need rhythm, grammar, context, and reassurance. Models trained on huge symbolic mixtures may not need all of that scaffolding every time.