他权倾朝野 却偏偏选一个弱智儿子做了太子！

Question

GenAI, which includes prompt engineering, is a subset of AI. Therefore, I wonder: What kind of questions are on-topic here but not already on-topic on AI.SE, DataScience.SE or Stats.SE?

I'm looking at existing questions here and they are either way too broad or already perfectly on topic on on AI.SE, DataScience.SE or Stats.SE (those 3 sites already greatly overlap between each other). And asking for "ready-to-go" prompts is off-topic here.

---

Examples:

• Does ChatGPT know how to count? posted here is similar to How does ChatGPT know math?,?How is GPT 4 able to solve math?, ?Why is ChatGPT bad at math?, ...
• What's the difference between the terms "ChatGPT", "GPT", "GPT-4", and "GPT-3.5-turbo"? posted here is similar to Are GPT-3.5 series models based on GPT-3?
• Are there powerful text generators that preserve attribution? posted here is similar to How can a language model keep track of the provenance of the main knowledge/sources used to generate a given output?
• How do I get leonardo.ai to add correct, legible text to images? posted here is similar to Why can't AI image generators output verbatim text when prompted to do so?
• How do I "teach" a large language model new knowledge? posted here is similar to How do you add knowledge to LLMs on CV.
• How important are GPUs vs. CPUs when training an LLM? could have been posted on DataScience.SE, e.g. see After the training phase, is it better to run neural networks on a GPU or CPU?
• etc.

Answer 1

The over-all impression I get, based on the types of questions I see on both sites at this time, is that GenAI SE is more focused on the needs of people wanting to use AI tools to generate content, or to solve some real-world problem in their lives using existing AI technology. In contrast, AI SE seems primarily concerned with questions programmers and developers are encountering when trying to build AI systems.

Naturally there is a considerable amount of inevitable overlap, but it seems that there is a purpose to having two sites, in that more technically oriented users won't need to sift through heaps of non-technical "how do I do this or that" type of questions. On the other hand, those with more end-user related questions won't need to pick through a bunch of questions that are too specialized to be relevant to them.

As for Data Science SE, it is clear that the questions are nearly all technical in nature, but lean toward the subject of the training data itself. Or for that matter, data that is being collected and processed for other purposes besides machine learning as well, including data visualization. Yes, to add further complexity, that links to questions with the dataviz tag on Cross Validated (stats.SE). (What's the difference between data science and data visualization in statistics? I will let the scientists in those fields debate that matter as they decide where to post.)

So, while a question could conceivably be a fit for multiple SE sites, there is probably a "best fit" in most cases. I, for one, am really happy there is GenAI now, because I can stop dropping non-technical questions into AI SE only to have them downvoted or ignored by the more technically focused crowd over there. But if I do have a legitimately technical AI question, I will post it in AI SE. That said, if the technical AI question is primarily about generative AI, there's no reason it wouldn't belong on GenAI SE. "Generative" is in the name, after all.

This answer is based on my observations and judgement so far. Anyone feel free to disagree with my take on the matter.