Re: ChatGPT > Spamassassin? :)

W dniu 2024-06-25 15:55, John Hardin napisał(a):
> On Mon, 24 Jun 2024, Mark London wrote:
> 
>> I received a spam email with the text below, that wasn't caught by 
>> Spamassasin (at least mine).   The text actually looks like something 
>> that was generated using ChatGPT.  In any event,  I put the text 
>> through ChatGPT, and asked if it looked like spam.  At the bottom of 
>> this email , is it's analysis.  I've not been fully reading this 
>> group.  Has there been any work to allow Spamassassin to use AI?  
>> Thanks.  - Mark
> 
> In a very limited manner. There is code in the repo that allows you to 
> set up ham and spam corpora and scan the spam corpora to pick out 
> common phrases and filter them via the ham corpora, then create 
> rules.based on the phrases and (IIRC) combinations of them.
> 
> This was being used to generate dynamic fraud rulesets (the "sought" 
> rules, still somewhat there as ADVANCE_FEE rules which I occasionally 
> manually update) until Justin Mason left the project. It's been 
> languishing since as he was providing the resources (infra and 
> maintenance) to run it for those rules. I was feeding those corpora for 
> a long time.
> 
> Take a look in the repo at the stuff under:
> 
>   https://svn.apache.org/viewvc/spamassassin/trunk/masses/rule-dev/
> 
>   
> https://svn.apache.org/viewvc/spamassassin/trunk/masses/evolve_metarule/
> 
> I don't know whether the project would be willing to set up infra to 
> revive dynamic advance fee fraud (or more general) rule generation, but 
> it's possible if someone was willing to bring that code up-to-date and 
> figure out what was needed and corpora providers were available.


This code still works, at least for me. I'm using my own corpora.