this is some very well-drafted corporate bullshit. the feature was enabled by default as an "opportunity", you see. from the screenshot the article has of Dropbox's actual UI: "internal" is an...
In a statement to Ars Technica, a Dropbox representative said, "The third-party AI toggle is only turned on to give all eligible customers the opportunity to view our new AI features and functionality, like Dropbox AI."
this is some very well-drafted corporate bullshit. the feature was enabled by default as an "opportunity", you see.
from the screenshot the article has of Dropbox's actual UI:
We only use technology partners we have vetted. Your data is never used to train their internal models, and is deleted from third-party servers within 30 days.
"internal" is an interesting qualifier there. I guess they can use customer data to train "non-internal" models?
Your data ... is deleted from third-party servers within 30 days.
again, masterful corporate bullshit.
they're getting you to agree to your data being sent to third-party servers, without explicitly saying that they send it, just that they delete it.
and within 30 days...of what? within 30 days of it being sent? or within 30 days of when the third-party stops using it?
this sort of thing is going to be the end state of a lot of tech companies. any data that can be mined for AI, will be mined for AI.
in particular, a lot of AI research is concerned about "model collapse" - the proliferation of AI-generated content on the internet being fed back in to subsequent training datasets. a corpus that is guaranteed to be mostly real-human-generated, such as "contents of a bunch of user dropboxes", is going to be incredibly valuable.
Last paragraph: turning it off is easy! I shouldn't have to turn it off. Not just dropbox but I hate this....all of this.....we have cloud computing and cloud storage and all this wonderful access...
Last paragraph: turning it off is easy!
I shouldn't have to turn it off.
Not just dropbox but I hate this....all of this.....we have cloud computing and cloud storage and all this wonderful access and off site backups and collaboration etc and.....it's all getting ruined by greed.
Would anyone use self storage depot that lets others sift through your stuff and take copies of your writing and photograph copies of your family portrait? "But it's easy to turn off drrr" doesn't cut it.
Just a random thought. Is it possible to crowd source a way to poison our own wells? Make a program that will randomly generate a terabyte of useless text full of grammatical errors and sentences...
Just a random thought. Is it possible to crowd source a way to poison our own wells?
Make a program that will randomly generate a terabyte of useless text full of grammatical errors and sentences of mad libs complete gibberish.
But -- every document it generates will contain a specific key phrase that only the owner knows, so that when the owner wants to search through their own stuff they can exclude garbage results.
Upload the terabyte of garbage somewhere for training AI software to pick up.
Share program with everyone and everyone start producing garbage effortlessly to obscure their private data as one leaf among a forest.
Quoting this useful comment by DCStone This does still spook me though. I don't have that much data on Dropbox...I guess I should start looking for an alternative.
Quoting this useful comment by DCStone
Didn't see it mentioned in the article, but per the linked FAQ it says the alpha AI applies to:
In countries with the preferred language set to English.
Excluding Canada, the UK (United Kingdom), and countries within the EEA (European Economic Area).
Which explains why I couldn't find this when I went into settings - bullet dodged (for now). Also looks like it's only for paid accounts so far?
This does still spook me though. I don't have that much data on Dropbox...I guess I should start looking for an alternative.
I've been very satisfied self-hosting Nextcloud (for years now). A modest technical hurdle up front, but, after that, it's just minor maintenance (updating regularly).
I've been very satisfied self-hosting Nextcloud (for years now). A modest technical hurdle up front, but, after that, it's just minor maintenance (updating regularly).
Yep, been doing the Owncloud/Nextcloud thing here for personal file storage as well as hosting files for family members who don't want to use cloud services.
Yep, been doing the Owncloud/Nextcloud thing here for personal file storage as well as hosting files for family members who don't want to use cloud services.
So it seems that countries that take cyber security a bit more seriously than the U.S.A. aren’t visibly able to be involved with Dropbox scraping “private” user data. Because, I’m assuming, it’s...
So it seems that countries that take cyber security a bit more seriously than the U.S.A. aren’t visibly able to be involved with Dropbox scraping “private” user data. Because, I’m assuming, it’s against their laws to do so. I realize that no data is really private if you don’t own what it’s stored on but doesn’t this beg the question as to why you would ever store anything in Dropbox if what they want to do with your data is illegal in many countries?
Companies that believe in cyber security Vs corporations allowed to sell your data as there isn't any country wide legislation to stop it. The difference between Windows 12 AI edition in the EU Vs...
Companies that believe in cyber security Vs corporations allowed to sell your data as there isn't any country wide legislation to stop it.
The difference between Windows 12 AI edition in the EU Vs US will be very interesting.
After having been a very big fan of Dropbox in its early years, I ditched it back when Condoleezza Rice joined their board. Genuinely had/have zero interest in storing my personal files on a...
After having been a very big fan of Dropbox in its early years, I ditched it back when Condoleezza Rice joined their board. Genuinely had/have zero interest in storing my personal files on a service with a name like that attached to it. At times I thought maybe I was being paranoid, but this kind of reaffirms my decision all those years ago.
This is disappointing. I've been using Dropbox for years. Since its early days. What service would be good for someone who isn't very technically inclined? I guess I'm not technologically...
This is disappointing. I've been using Dropbox for years. Since its early days.
What service would be good for someone who isn't very technically inclined? I guess I'm not technologically illiterate, but more busy and tired with very little mental capacity left.
I do not want to use Google Drive or One Drive though.
I personally use Resilio Sync (or Synthing if you prefer FOSS), which is a P2P syncing solution. I have a bunch of folders that are synced between my PC, phone, and Steam Deck. Advantages: easy to...
I personally use Resilio Sync (or Synthing if you prefer FOSS), which is a P2P syncing solution. I have a bunch of folders that are synced between my PC, phone, and Steam Deck.
Advantages: easy to setup (Resilio is more user-friendly than Syncthing IMO), you don't rely on an external 3rd party to host your files
Disadvantages: your files can only be synced if 2 devices are up and running at the same time, no web interface
To workaround the 2 disadvantages, it requires a bit more tech knowledge:
My most "important" folders are also synced on my VPS, so they're always up-to-date
I installed Filebrowser on top of those folders so that I have a web interface to quickly drop files and fetch files when needed.
It's definitely not a plug-and-play solution, but it's very low maintenance once installed.
A Virtual Private Server. A computer, connected to the internet, that you rent through some provider. You can use them for anything, usually to host websites and services. I use the cheapest...
A Virtual Private Server. A computer, connected to the internet, that you rent through some provider. You can use them for anything, usually to host websites and services.
I use the cheapest option from Hetzner, and they're scalable to your needs (more CPU, more RAM, more disk).
So I don’t personally understand the fears people have of their data being used to train an LLM (I’d love to hear from people that feel this way). But I do think it’s incredibly stupid to take a...
So I don’t personally understand the fears people have of their data being used to train an LLM (I’d love to hear from people that feel this way). But I do think it’s incredibly stupid to take a product sold as a secure place to store files and start littering those files across 3rd party servers.
Personally, it's not fear, it's a refusal to let companies take my data, my private files, to train their LLM models with it to make a profit on my back. And also yes, there's an expectation of...
Personally, it's not fear, it's a refusal to let companies take my data, my private files, to train their LLM models with it to make a profit on my back.
And also yes, there's an expectation of privacy. Even if it's written somewhere in their neverending and user-hostile ToS, you can't reasonably expect your average user to be cool with the idea that private and sensitive documents are just feed in a machine whose goal is to regurgitate them to anyone.
I don't trust the AI companies to properly protect private data. Their track record isn't the best. They don't seem to have full control of their models with how they frequently get "hacked" and...
I don't trust the AI companies to properly protect private data. Their track record isn't the best. They don't seem to have full control of their models with how they frequently get "hacked" and possibly tricked into revealing training data
I'm not sure I fall into the fear camp as it doesn't sound like this is for straight up training a language model for conversation. However, after seeing an article recently that showed that...
I'm not sure I fall into the fear camp as it doesn't sound like this is for straight up training a language model for conversation. However, after seeing an article recently that showed that chatgpt could be coerced into spitting out what appeared to be full strings of training data (like email addresses and phone numbers), I think it's reasonable to be weary about your person documents being fed into a potentially public facing model. Just because the model spits out "unique" outputs for a prompt under normal use cases doesn't mean that the original inputs aren't still accessible under the surface if you start pushing on edge cases. I'm not sure I'd want things like personal medical records or tax documents being fed into the same model other people can interface with.
this is some very well-drafted corporate bullshit. the feature was enabled by default as an "opportunity", you see.
from the screenshot the article has of Dropbox's actual UI:
"internal" is an interesting qualifier there. I guess they can use customer data to train "non-internal" models?
again, masterful corporate bullshit.
they're getting you to agree to your data being sent to third-party servers, without explicitly saying that they send it, just that they delete it.
and within 30 days...of what? within 30 days of it being sent? or within 30 days of when the third-party stops using it?
this sort of thing is going to be the end state of a lot of tech companies. any data that can be mined for AI, will be mined for AI.
in particular, a lot of AI research is concerned about "model collapse" - the proliferation of AI-generated content on the internet being fed back in to subsequent training datasets. a corpus that is guaranteed to be mostly real-human-generated, such as "contents of a bunch of user dropboxes", is going to be incredibly valuable.
Last paragraph: turning it off is easy!
I shouldn't have to turn it off.
Not just dropbox but I hate this....all of this.....we have cloud computing and cloud storage and all this wonderful access and off site backups and collaboration etc and.....it's all getting ruined by greed.
Would anyone use self storage depot that lets others sift through your stuff and take copies of your writing and photograph copies of your family portrait? "But it's easy to turn off drrr" doesn't cut it.
Just a random thought. Is it possible to crowd source a way to poison our own wells?
Make a program that will randomly generate a terabyte of useless text full of grammatical errors and sentences of mad libs complete gibberish.
But -- every document it generates will contain a specific key phrase that only the owner knows, so that when the owner wants to search through their own stuff they can exclude garbage results.
Upload the terabyte of garbage somewhere for training AI software to pick up.
Share program with everyone and everyone start producing garbage effortlessly to obscure their private data as one leaf among a forest.
(Edit for formatting)
Quoting this useful comment by DCStone
This does still spook me though. I don't have that much data on Dropbox...I guess I should start looking for an alternative.
I've been very satisfied self-hosting Nextcloud (for years now). A modest technical hurdle up front, but, after that, it's just minor maintenance (updating regularly).
Yep, been doing the Owncloud/Nextcloud thing here for personal file storage as well as hosting files for family members who don't want to use cloud services.
So it seems that countries that take cyber security a bit more seriously than the U.S.A. aren’t visibly able to be involved with Dropbox scraping “private” user data. Because, I’m assuming, it’s against their laws to do so. I realize that no data is really private if you don’t own what it’s stored on but doesn’t this beg the question as to why you would ever store anything in Dropbox if what they want to do with your data is illegal in many countries?
Companies that believe in cyber security Vs corporations allowed to sell your data as there isn't any country wide legislation to stop it.
The difference between Windows 12 AI edition in the EU Vs US will be very interesting.
I switched to pcloud drive a while back and have been very satisfied, fwiw.
Lovely, my old personal journal .txt files will be used to train the next AI model. What an amazing "opportunity"...
That's not going to give AI a bias in the slightest, eh?
After having been a very big fan of Dropbox in its early years, I ditched it back when Condoleezza Rice joined their board. Genuinely had/have zero interest in storing my personal files on a service with a name like that attached to it. At times I thought maybe I was being paranoid, but this kind of reaffirms my decision all those years ago.
This is disappointing. I've been using Dropbox for years. Since its early days.
What service would be good for someone who isn't very technically inclined? I guess I'm not technologically illiterate, but more busy and tired with very little mental capacity left.
I do not want to use Google Drive or One Drive though.
I personally use Resilio Sync (or Synthing if you prefer FOSS), which is a P2P syncing solution. I have a bunch of folders that are synced between my PC, phone, and Steam Deck.
To workaround the 2 disadvantages, it requires a bit more tech knowledge:
It's definitely not a plug-and-play solution, but it's very low maintenance once installed.
Thank you for that! I'll look into it.
What is VPS?
A Virtual Private Server. A computer, connected to the internet, that you rent through some provider. You can use them for anything, usually to host websites and services.
I use the cheapest option from Hetzner, and they're scalable to your needs (more CPU, more RAM, more disk).
If you do want an actual cloud storage solution, rather than something P2P, I use pcloud drive and have been very pleased with them.
So I don’t personally understand the fears people have of their data being used to train an LLM (I’d love to hear from people that feel this way). But I do think it’s incredibly stupid to take a product sold as a secure place to store files and start littering those files across 3rd party servers.
Personally, it's not fear, it's a refusal to let companies take my data, my private files, to train their LLM models with it to make a profit on my back.
And also yes, there's an expectation of privacy. Even if it's written somewhere in their neverending and user-hostile ToS, you can't reasonably expect your average user to be cool with the idea that private and sensitive documents are just feed in a machine whose goal is to regurgitate them to anyone.
I don't trust the AI companies to properly protect private data. Their track record isn't the best. They don't seem to have full control of their models with how they frequently get "hacked" and possibly tricked into revealing training data
I'm not sure I fall into the fear camp as it doesn't sound like this is for straight up training a language model for conversation. However, after seeing an article recently that showed that chatgpt could be coerced into spitting out what appeared to be full strings of training data (like email addresses and phone numbers), I think it's reasonable to be weary about your person documents being fed into a potentially public facing model. Just because the model spits out "unique" outputs for a prompt under normal use cases doesn't mean that the original inputs aren't still accessible under the surface if you start pushing on edge cases. I'm not sure I'd want things like personal medical records or tax documents being fed into the same model other people can interface with.
Coincidentally, I recently canceled my sub with them and began migrating all my storage to Proton Drive. I couldn't be happier!
Thanks for the heads up about this. Only use dropbox free to backup certain older digital data. Really alarming how its enabled by default 😅.