A new Apple preprint has appeared on Arxiv.
https://arxiv.org/pdf/2403.20329.pdf
Reference resolution is an important problem, one that is essential to understand and success- fully handle context of different kinds……We also bench- mark against GPT-3.5 and GPT-4, with our smallest model achieving performance com- parable to that of GPT-4, and our larger models substantially outperforming it.
Looks like more work towards on device LLM.