We can all agree that the new Qwen models are truly amazing, and we are blessed to have them. In coding, they are certainly a breakthrough.
However, lately as I've been working on my app's App Store copy and screenshots, I've been thinking that this is something that they don't necessarily excel at. Compared to Sonnet 4.6, they are still considerably behind, and don't really understand the deep semantic connections that are required for such a task. What models do you guys use for such tasks?
Also, another thing that I would truly love to see is models with the conversational intuition of Claude models. I can't stand how almost every model just tries to talk as much as they can for every query, and seemingly only Anthropic figured out how to make a model answer just as much as needed, and even proactively ask clarifying questions. I was thinking that maybe this would be easier to fix with a finetune (I remember seeing Qwopus finetunes a month or so ago), but these usually messed too much with the chain-of-thought, degrading overall quality.
What are your thoughts?
[link] [comments]