3 ms·

I wonder. People definitely talk differently online. Is it possible for someone to filter their online presence enough to not be able to glean much of a diagnos

by ianai 3y ago

I wonder. People definitely talk differently online. Is it possible for someone to filter their online presence enough to not be able to glean much of a diagnosis? Does seem interesting.

PaulHoule 3y ago
The easy way to evade a system like that is the same way spammers beat Spam Assassin and other open-source spam filters. Run your text through the analysis model and if you don't like what the model says, change the text. I am filtering more and more of my social media participation through my intelligent agent YOShInOn, particularly all the links I post to HN and Mastodon are selected by the two of us working together. People have characterized it as "random but never hostile" and I'm pretty proud of that. I definitely like YOShInOn have my back in enforcing a brand image and the way things are going YOShInOn will be able to speak in my name when I no longer can.
- ianai 3y ago
  Is that only available on FB? I'm not getting much from google and have no FB. Good idea though.
  - PaulHoule 3y ago
    YOShInOn is (mostly) an RSS reader, I have it pointed at 93 RSS feeds right now so I have a $9.30 cent monthly bill for Feedburner. I get a lot of interest from people who would like a similar product, it is something I think about productizing but I am not ready to quit my day job. It has the problem that it takes 1000 or so judgements to build a good model of your preferences, I am sure with effort it could be improved somewhat in the “few shot” case but since I am swimming in judgements I don’t have any personal motivation to do so. If i were going to make a cloud version of it I’d end up doing collaborative filtering anyway because it would do better with “few shot”. An open source version is also interesting but it involves stopping what I am doing to make it better for me to make it something just anybody can install, in the long term I am modularizing it to make it possible but it competes with other side projects.

by ianai 3y ago

I wonder. People definitely talk differently online. Is it possible for someone to filter their online presence enough to not be able to glean much of a diagnosis? Does seem interesting.

PaulHoule 3y ago

The easy way to evade a system like that is the same way spammers beat Spam Assassin and other open-source spam filters. Run your text through the analysis model and if you don't like what the model says, change the text. I am filtering more and more of my social media participation through my intelligent agent YOShInOn, particularly all the links I post to HN and Mastodon are selected by the two of us working together. People have characterized it as "random but never hostile" and I'm pretty proud of that. I definitely like YOShInOn have my back in enforcing a brand image and the way things are going YOShInOn will be able to speak in my name when I no longer can.

ianai 3y ago
Is that only available on FB? I'm not getting much from google and have no FB. Good idea though.
- PaulHoule 3y ago
  YOShInOn is (mostly) an RSS reader, I have it pointed at 93 RSS feeds right now so I have a $9.30 cent monthly bill for Feedburner. I get a lot of interest from people who would like a similar product, it is something I think about productizing but I am not ready to quit my day job. It has the problem that it takes 1000 or so judgements to build a good model of your preferences, I am sure with effort it could be improved somewhat in the “few shot” case but since I am swimming in judgements I don’t have any personal motivation to do so. If i were going to make a cloud version of it I’d end up doing collaborative filtering anyway because it would do better with “few shot”. An open source version is also interesting but it involves stopping what I am doing to make it better for me to make it something just anybody can install, in the long term I am modularizing it to make it possible but it competes with other side projects.