Welcome to Import AI, a newsletter about AI research. Import AI runs on arXiv and feedback from readers. If you’d like to support this, please subscribe. Subscribe now Google paper suggests that LLMs simulate multiple personalities to answer questions:…The smarter we make language models, the more they tend towards building and manipulating rich, multi-agent world…

Import AI 

Recent AI research reveals LLMs simulate multi-agent "societies of thought" when reasoning through complex problems, with models like DeepSeek-R1 exhibiting distinct personas and debate patterns. ChipBench benchmark shows frontier models struggle with real-world Verilog chip design, achieving only 22-50% success rates on practical tasks. Google's Gemini solved 2 novel Erdős math problems autonomously from 700 candidates, though human filtering was essential to identify meaningful solutions. Huawei developed AscendCraft to automate kernel generation for their chips using LLM-based DSL transcompilation, achieving 98% compilation success.

Import AI 444: LLM societies; Huawei makes kernels with AI; ChipBench