Path: utzoo!mnetor!uunet!lll-winken!lll-lcc!ames!pasteur!agate!garnet!weemba
From: weemba@garnet.berkeley.edu (Obnoxious Emacs Weenie)
Newsgroups: news.misc
Subject: Re: Take a sniff of Gnews
Message-ID: <7993@agate.BERKELEY.EDU>
Date: 25 Mar 88 04:59:44 GMT
References: <7928@agate.BERKELEY.EDU> <25590@yale-celray.yale.UUCP> <40@ncar.ucar.edu> <7961@agate.BERKELEY.EDU> <47@ncar.ucar.edu>
Sender: usenet@agate.BERKELEY.EDU
Reply-To: weemba@garnet.berkeley.edu (a small gnews-ance)
Organization: Brahms Gang Posting Central
Lines: 113
In-reply-to: woods@ncar.ucar.edu (Greg Woods)


In article <47@ncar.ucar.edu>, woods@ncar (Greg Woods) writes:

>>	        I'm not trying to write any definitive rules right now.
>
>   But sooner or later you have to.

Not me.  But that I can get very close to the "right" rule about talk.ab-
ortion on my second try, without trying very hard, is indicative that the
writing of some such rules isn't going to be hard.  Obviously they can't
cover every situation--that *would* required impossibly advanced AI--and
just as obvious no one is proposing that these rules cover every situa-
tion.

>				     My point is that I do not believe that
>you can do so in any reliable way. Convince me I'm wrong (I am keeping an
>open mind) and I'll believe your proposal is a "good thing".

I'll let someone else do the convincing.  If a few people are seriously
interested in working this out, they'd presumably exchange "fairy stories"
and similar rules, and then try to knock each other's suggestions down,
until they reach some kind of stability.  Then they'd try to convince you.

>>The user is free to answer "no" to the "do we understand each other?"
>>question, ie, "No, Mr Computer Sir, you didn't understand me.  Shut up
>>and let me post."
>
>   Look at how many complaints we already get about the 50% rule.

And in retrospect it should have been experimented with a little more
heavily before it was installed net-wide.  I'm suggesting this very
proposal to those interested in writing rules-based screening for news
posting.

>								    This is
>more of the same, except that the heuristics are much more difficult to
>define in any meaningful way.

Exactly.  Whoever wants to write such a system has to work hard at it.

>>I see that you are not rebutting these [fairy] stories, but are pointing out
>>an external difficulty.  Is this a concession, then, that you agree that such
>>fairy stories, if implemented, would have a beneficial effect?
>
>  Well, there were over 100 lines of "fairy stories" and I only have
>so much time to read news :-) I personally think they might be beneficial,

Now how does this comment about "beneficial" square with your above comments
about "reliability" and "much more difficult"?  They didn't take much effort
to think up.

(Really, Greg, if you're trying to save time, then just agree with me!)

>but only if they are kept up-to-date.

So when will discouraging posted votes in news.groups go out of date??

>				       I really don't like the idea of hard-
>coding things for individual newsgroups into the posting software. It
>makes it much harder to change the newsgroup structure (including such
>things as renaming groups or deleting groups) and I think we need to
>retain maximum flexibility in this area.

You would be absolutely correct, if we were trying to set up a sophisti-
cated system, sniffing around for all possible violations of the "rules".

But something much more low-key, dealing mostly with the things that long
net.experience has determined are ungood--eg, vote-posting in news.groups
--is based on what's been with the net for a long time, and will probably
not go away, not become obsolete, and not need much in the way of mainta-
nance.

Whoever writes this kind of system has a lot of hard thinking to do.  But
I don't think he needs to do any advanced AI, which is what I was object-
ing to when I first got into this discussion.

>>And there's no point in trying it out if the re-
>>action from the backbone is a loud "hell no".  Remember junker?

>  Unless I explicitly say so, these are just my PERSONAL OPINIONS and do
>NOT, repeat NOT, represent any official position of the backbone.  [...]

>  And as for "hell no", I'm not ruling anything out completely. I just have
>to be convinced it will work and do some good before I would advocate having
>it included in the official news software distribution.

Exactly my point.  Whoever wants to write something like Brad suggested needs
to know ahead of time if there's going to be serious, permanent trouble or if
he merely needs to be slick with his programming and marketing skills.

>  All sorts of reasons, including such things as frequency and magnitude
>of changes. One reason people can get away with maintaining news as a 10th
>priority is that the software is stable and requires relatively few changes.

Since I envision that only things that long experience has agreed are a waste
of net.bandwidth, I don't see why it'll be more than a one-time change, and
not distributed separately but folded in with some later major news software
release.

>  Naturally, if you want to try something like this out on your own site,
>no one is stopping you. In general, I am against keyword-based news systems
>because they sound great in theory, but I don't believe they will work in
>practice.
b
Keyword-based news system try to classify everything, and must expand to
keep up with new topics or die.  These screening suggestions are just try-
ing to classify a few already well-known things.  All your objections seem
to be against the former.

>	   You're right in saying I don't KNOW they won't work, but you don't
>know they WILL.

Which is one of the reasons that I am not going to be writing such a system.

ucbvax!garnet!weemba	Matthew P Wiener/Brahms Gang/Berkeley CA 94720