Is using AI to improve and refine responses considered a mistake on this forum?
Yes, because typical AI can change the content so much, that it is not just a grammar correction tool. It can change whole words and sentences, and quite often hallucinate, and produce responses, which looks correctly, but are totally wrong, when you dig into the details.
I believe the main goal is to provide relevant and accurate information.
AI can only provide you some data, if they were in the training set, or if a given set of algorithms, used to made a given AI, are smart enough to produce it out of thin air. In all other cases, models will hallucinate, because they are designed to do so. If you have an answer for f(0) and f(1) in a given training set, then the answer for f(0.5) is a mix of both. If you have a big training data, and proper algorithms, then many hallucinations are correct, so AI is perceived as "intelligent". But it isn't much smarter, than data and algorithms, which were used to make it.
Considering the increasing use of AI in the digital world, is it reasonable to completely ban its use on this forum?
Yes, as long as people will copy-paste all answers from bots, without thinking about it. If I want to talk with bots, then I have a lot of tools, where I am sure, that all replies are produced automatically. I can even run some offline models, to be 100% sure, that I am talking with AI, and not with thousands of humans, connected to some "smart network" in "AI startup".
Here, forum is used mainly by humans, and many people prefer human-written answers. And as long as AI answers are below some level of correctness, it should stay that way. And any AI copy-pasted posts should be reported as plagiarism, especially when AI companies are scraping the web, and mixing content with different licenses, while violating all copyrighted materials. Either that, or "intellectual property" should be discarded entirely as a concept, no matter if it is done by humans or bots. Because if bots can plagiarize any content, then why humans shouldn't. And if nobody should, then bots should also be trained only with data, where the AI creator has the right to use and distribute, and where no license restricts that.
AI can provide useful and relevant information, as long as the results are carefully filtered.
You can get out of AI only things, which were put into AI during training.
Could this forum consider a similar feature, rather than completely banning AI use?
AI usage is not banned. The text you are reading now, is written by humans. However, it is not a product of a single individual, but a collective effort of at least three people, if not more (I am not sure about the total size of our network, I can only see shared public keys, and my job is to collect pseudonymous messages, and publish it, with a random time delay, if I consider it useful). We use different tools to produce our content, but the final version is always read by a real human, at least once before publishing, and we use automated tools only as a source of information, but not the source of content. None of my posts are auto-generated, they are always human-made, but there are many tools in use, which can provide accurate calculations (and for example not hallucinate, when I share SHA-256 of something, ECDSA calculations, signed transaction data, or anything else).
Edit:
A prime example of incorrect AI 'answers' is it all too often calling mining "a process of solving complex mathematical equations".
This. Ask about any hash function, and AI will tell you, that it is "complex". While in practice, it can be implemented with just a few hundred lines of code in C++, as done by vjudeu here:
https://bitcointalk.org/index.php?topic=293382.msg61865839#msg61865839And then, you can copy-paste it here, and explore it, try to change values, constants, break some rounds, and so on:
https://www.programiz.com/cpp-programming/online-compilerOr, you can have a more copy-pasterino version, for exploring each individual value of SHA-256. My implementation:
https://github.com/stwenhao/small_curves/blob/master/sha256.cppMy version could be much shorter in practice, but all loops are expanded on purpose, to make analysis easier, and to allow focusing on individual rounds, and writing equations like "z21=g20+s2(h24)+ch(h24,e20,f20)+k21+w21" easier, and compare, if given simplifications are true or not, for a given message.
