Invited Talk 1 by Abdelrahman Mahmoud (Assistant Professor at MBZUAI): Speaking in Tongues: Binary, Language Models, and Neural Signals

Large Language Models (LLMs) have forced us to rethink what it means for machines to understand and generate language. But the most interesting frontier may not be natural language at all; it may be the many non-human languages that quietly structure modern computing! In this talk, I explore this broader landscape of machine “tongues,” where meaning is encoded not in words and sentences, but in opcodes, register conventions, and architectural constraints.

While LLMs are often applied to text, I will show that their fluency can extend to an (unnatural?) domain: raw assembly code. A central part of the talk reveals how LLMs can interpret and translate directly between Instruction Set Architectures, treating x86, ARM, and RISC-V not as opaque binary formats, but as distinct dialects of a shared computational language. This reframes assembly-to-assembly translation as a linguistic and representational task, rather than merely a compiler engineering problem, and opens up new possibilities for portability and understanding in settings where source code is absent.

By examining these unexpected capabilities, I invite the audience to reconsider what it means for machines to “speak” and how we can leverage their fluency across the diverse languages of computation. Speaking in Tongues highlights both practical and conceptual opportunities: from rethinking how software moves across architectures to imagining a broader landscape in which models navigate the many structured languages embedded in our technological world.