Posted by usefulposter 2 days ago
Unless you're a billionaire* or a CEO firing off memos where you fire half your company's workforce.
u got to be powerful to puond out a txt this way and have ppl still listen to u.
Otherwise, it is getting dismissed because 'you didn't put enough effort into the comment, so I'm not going to read it.'
That is amusing to me.
*Reference to the analysis performed on the Epstein emails and texts.
So develop and fund and use AI but manually paraphrase things and don't cite AI?
It is best to cite a source and/or a method.
Do you think it is better to paraphrase and not cite AI?
I don't recall encountering posts on HN that I've wanted to flag as AI.
Have you considered this?
If people do not cite their sources or methods when they use AI, then we will not know where error was introduced by paraphrasing AI.
If they say "no" to "did you use AI", they're probably not correct and/or lying.
But you may not cite or quote or link to AI generated work?
> If people do not cite their sources or methods when they use AI, then we will not know where error was introduced by paraphrasing AI.
1. No AI comments without a human in the loop.
2. Please cite. Please cite when you use AI so that others can trace the errors and evaluate the premises of the argument. An argument has premises and a logical form.
We should expect the frequency of AI errors like hallucinations to decrease and accuracy to increase over time.
You should always consider peer review and getting another opinion regardless of whether AI or ML were used.
Do you need to cite AI?
If scientific reproducibility is necessary or important for your application,
You should also cite search queries, search results at that time, the name and version and software package hash of each software tool, the configuration parameters for each software tool, the URL and hash of the data, and whether you used spell check or autocorrect or an AI grammar service.
If you use an (AI) grammar service, you should disclose the model name and version, model hash or Merkle hash, and the model parameters.
But most people don't even cite URLs here; it's just people making unsupported arguments.