CFP: RustINDIA Conference 2026

Abstract

Now that LLMs are capable of generating working code in pretty much any programming language, it raises the question of which language to use? Is an LLM equally good at generating safe C code versus Rust? Which language helps LLMs correct themselves the best?

Previously, I’d presented the shortcomings of using LLMs for Systems programming at the FSTTCS conference. Models from OpenAI, Anthropic and Google had a tendency to produce C code that looked correct, but had crash rates approaching ~30% for Sanitizers (ASAN, UBSAN, INTSAN). Allowing the model to learn from the feedback of the sanitizers, varying the temperature, the kind of prompt and giving secure examples didnt help much. What did help was changing the language - Rust had far fewer runtime crashes.

In this talk I demonstrate the results of a new series of experiments to show the impact of using Rust over more traditional languages like C. I explore projects like Anthropic’s CCC, and try to push LLMs to their limit.

Biography

Asutosh Pandey is a Compiler Engineer at AMD working with the CPU Compiler team. He co-organizes the LLVM Social Bangalore Meetup, the Innovations In Compiler Technology Workshop and the Segfault Hackathon.

LinkedIn: https://www.linkedin.com/in/ashupdsce

Twitter: https://x.com/ashpande18

P.S : Possible to have 10 minutes for the lightning talk instead of 5? That would be the sweet spot.