This isn’t really true (or at least, doesn’t apply across the board). Qwen (Alibaba’s open source model) outperforms GPT4 on Chinese language tasks, and I can f

This isn’t really true (or at least, doesn’t apply across the board). Qwen (Alibaba’s open source model) outperforms GPT4 on Chinese language tasks, and I can further finetune it for my own tasks (which I’ve done, and I confirm it’s produces more natural output than GPT4). Other benchmarks/anecdotes suggest fine-tuned code models are outperforming GPT4 too. The trend seems to be that smaller, fine-tuned task specific models outperform larger generalised models. It requires a lot of resources to pretrain the base model, but as we’ve seen, there’s no shortage of companies who are willing and able to do that. Not to mention, all those other companies are already profitable, whereas OpenAI is already burning investor cash.