[extropy-chat] META: Anti-Bayesian spam
pharos at gmail.com
Tue Dec 12 21:10:23 UTC 2006
On 12/6/06, Robert Bradbury wrote:
> Well, the following got through the gmail filters this morning. I've
> converted it to text and removed the GIF attachment (which was
> presumably the real message). Interesting how it seems to be
> generating relatively coherent text. I wonder if they are picking out
> phrases from popular news items or simply generating semi-proper text
> using phrases that score low on Bayesian filtering methods.
> At any rate the battle continues...
Just read an article on ZDNet that reports that many other people are
noticing this new type of spam.
December 11th, 2006
If this "personalized spam" can scale, we and the Net are in very big trouble
Posted by David Berlind @ 11:55 am
So, why write about spam again now? Because I've noticed a new kind of
spam that's quite amazing in how it can probably beat most anti-spam
systems and I'm wondering if robots are doing the work or if some
human is doing this by hand. I suspect it's robots which is why it has
me worried because it looks like there are humans at work (not a
scalable system). This new form of spam appears to be very
personalized to me. For example, the subject lines in a lot of the
spam I've been getting appears extremely targeted to my interests.
One talkback post on this article suggested
"David, the first example you gave is intended to ruin the utility of
Bayesian filters. Most likely, the scraper put together "sightings" of
your address and correlated it with others found nearby, then
generated word salad with words or phrases common to those locations."
The good news is that the gmail spam filters are still very good at
stopping almost all spam. The other good news is that gmail blocks all
images unless you click on 'Always allow images from this sender'.
Even more good news is that gmail attachments have to be deliberately
downloaded by you.
So thumbs up to gmail!
More information about the extropy-chat