What can an LLM build with only 32 CPU instructions? (github.com)

Picos has introduced a groundbreaking experiment involving a custom 32-bit CPU, dubbed smol32, to test the capabilities of large language models (LLMs) under stringent constraints. The setup includes fixed RAM, basic storage, and a simple 256 × 192 pixel display, with no operating system or conventional programming framework. This environment forces models to construct applications entirely from raw machine instructions, limited to just 32 operations ranging from basic arithmetic to device I/O. Users can customize their experiments by adjusting CPU speed, memory, and disk space, all while monitoring the LLM's resource usage in real-time. This initiative is significant for the AI/ML community as it explores how much complexity and functionality LLMs can build autonomously when stripped of traditional computational supports. By limiting available tools and enforcing strict execution boundaries, Picos aims to assess the creativity and resourcefulness of LLMs in developing not only primitive functions but potentially more complex systems like shells or self-hosted compilers. The implications could lead to deeper insights into the adaptability of AI systems under severe constraints, paving the way for innovations in AI design and efficiency.