I remember that too. Vaguely. The trick was to put a delay in sufficient to get the colors solid @ 50hz? There was lots of cool things like that. For example the 3 voice sound emulation in Jet Set Willy. Happy memories of simpler times.
I wrote 2 nasty grams... (even this was difficult as the turbox form errored. There was a "If you are having problems go here" generic form link; however, it must have been designed by the the brainiacs-r-us dept because it sent me back to the broken form when I told it I was writing about Turbo Tax).
Annoy beyond belief I called sales and told them I wanted my money back. He said, OK send back the CD. I said, No way. You sent it to me unsolicited and I am NOT paying a cent to get it back to you. He put me on hold for a short eternity and camed back saying I was all set. Refund came thru a few days later.
I downloaded TaxCut and am very happy with it. Intuit are on my shit list permanently.
1. Nonsense. My personal behavior has nothing to do with the spam I receive. I am pleased about this, because the reverse is the goal of every spammer today. Spam is spam. It's not a case of "I'll know it when I see it". By the way, this model, if it is such an personalized description of your anti-taste, would be helpful to marketers long term I think, don't you?
2. I never delete the message completely. I prefer to keep a copy and email myself a a quick list of hits to review to make sure I got spam only, with URLs to each message.
When you say "it does the work" of building the taxonomy, what do you mean? The model must be trained and that requires cases does it not? I think the "work" will be the user flagging what is spam (and what is not is also important).
There are not a lot of false positives, but there are some.
There is a list of addresses that always get through. This is my basically my address book and manual additions, like mailing lists. If I get mail from one of these it always gets through whether it fails all other rules, BCC'ed etc.
After that I have a series of rules. I have been running my little server app for about 15 months and below I posted the recent stats. I don't count false negatives (spam that gets through) but doubt I have seen more than 20 messages in this time.
90 Invalid or unknown FROM domain 697 Body contains spam word(s). 263 Subject contains spam word(s). 162 From address ends in numbers 434 Address is malformed in some way 108 Subject ends in numbers 443 No recipients found. 7 Blocked sender found. 1798 No Valid Address Found. 13 From address is blank. 4015 Total
I always see a summary of what is nuked and keep a URL to the message so I can get it if necessary. Takes a few seconds to review this. Not perfect but my problem is solved to my satisifaction.
I don't think Bayesian is great at taxonomy either. Only asking for many node values to infer what is basically a binary result (spam or not). I think they are making it complicated if the attempt is to score the message for belief that it may be spam. To what end?
I would like to understand the choice of Bayesian more. As far as I know Bayesian is good for classifying based on *belief* and can be pretty good when only partial evidence is available to network. This is great for Marketing activities, eg sending out mass emails to a segment of a database:) . However as this is _my_ email and mission critical to me, just a simple belief that something is spam is not enough
In my experience, 99% of spam can be caught with static rules (am I in the TO or CC line gets a bit under half the spam I receive). Taxonomical analysis of the subject and body can get the rest.
Bayesian seems like overkill, or maybe even a bad fit. Let's face it, the other well known use for Bayesian is the famous Microsoft Office Paper Clip!!! And that is about as useful as the proverbial ashtray on a motorbike!!
Slashdot Mirror
I remember that too. Vaguely. The trick was to put a delay in sufficient to get the colors solid @ 50hz? There was lots of cool things like that. For example the 3 voice sound emulation in Jet Set Willy. Happy memories of simpler times.
I wrote 2 nasty grams ... (even this was difficult as the turbox form errored. There was a "If you are having problems go here" generic form link; however, it must have been designed by the the brainiacs-r-us dept because it sent me back to the broken form when I told it I was writing about Turbo Tax).
Annoy beyond belief I called sales and told them I wanted my money back. He said, OK send back the CD. I said, No way. You sent it to me unsolicited and I am NOT paying a cent to get it back to you. He put me on hold for a short eternity and camed back saying I was all set. Refund came thru a few days later.
I downloaded TaxCut and am very happy with it. Intuit are on my shit list permanently.
1. Nonsense. My personal behavior has nothing to do with the spam I receive. I am pleased about this, because the reverse is the goal of every spammer today. Spam is spam. It's not a case of "I'll know it when I see it". By the way, this model, if it is such an personalized description of your anti-taste, would be helpful to marketers long term I think, don't you?
2. I never delete the message completely. I prefer to keep a copy and email myself a a quick list of hits to review to make sure I got spam only, with URLs to each message.
When you say "it does the work" of building the taxonomy, what do you mean? The model must be trained and that requires cases does it not? I think the "work" will be the user flagging what is spam (and what is not is also important).
There are not a lot of false positives, but there are some.
There is a list of addresses that always get through. This is my basically my address book and manual additions, like mailing lists. If I get mail from one of these it always gets through whether it fails all other rules, BCC'ed etc.
After that I have a series of rules. I have been running my little server app for about 15 months and below I posted the recent stats. I don't count false negatives (spam that gets through) but doubt I have seen more than 20 messages in this time.
90 Invalid or unknown FROM domain
697 Body contains spam word(s).
263 Subject contains spam word(s).
162 From address ends in numbers
434 Address is malformed in some way
108 Subject ends in numbers
443 No recipients found.
7 Blocked sender found.
1798 No Valid Address Found.
13 From address is blank.
4015 Total
I always see a summary of what is nuked and keep a URL to the message so I can get it if necessary. Takes a few seconds to review this. Not perfect but my problem is solved to my satisifaction.
I don't think Bayesian is great at taxonomy either. Only asking for many node values to infer what is basically a binary result (spam or not). I think they are making it complicated if the attempt is to score the message for belief that it may be spam. To what end?
I would like to understand the choice of Bayesian more. As far as I know Bayesian is good for classifying based on *belief* and can be pretty good when only partial evidence is available to network. This is great for Marketing activities, eg sending out mass emails to a segment of a database :) . However as this is _my_ email and mission critical to me, just a simple belief that something is spam is not enough
In my experience, 99% of spam can be caught with static rules (am I in the TO or CC line gets a bit under half the spam I receive). Taxonomical analysis of the subject and body can get the rest.
Bayesian seems like overkill, or maybe even a bad fit. Let's face it, the other well known use for Bayesian is the famous Microsoft Office Paper Clip!!! And that is about as useful as the proverbial ashtray on a motorbike!!
Gary