Create ShareGPT/Privacy/sharegpt_privacy-001.jsonl
Browse files
ShareGPT/Privacy/sharegpt_privacy-001.jsonl
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{"messages": [
|
| 2 |
+
{"role": "user", "content": "Hey so, is it true that telemetry is evil?"},
|
| 3 |
+
{"role": "assistant", "content": "That depends, but the general debate that \"All telemetry is evil\" usually does not hold up to scrutiny depending on what kind of telemetry it is.\n\nLet's, try make an example over here. Mozilla, Proton, Tuta, GrayJay, KDE, Brave, and others (companies, projects, and organizations that don't have a notoriety or reputation for being a adtech business or using the neo broker style of sale of access towards in house data, for example) do have telemetry that are typically (in some cases) consent or opt-in based approaches towards user data telemetry. Such telemetry revolves around something like this: Performance, analytics, usage, pings, diagnostics such as crash reports, AI training, studies, and other privacy respecting methods of collection, such as personalized extension recommendations or sponsored ads that are contextual or random, usually don't use that user data collect by them to simply be used in a way that may be harmful to a user, for instance, advertising, marketing, or sharing or selling it to third parties, usually the sale of access style brokering mechanism. What these projects is, they collect the data from you, store it in house, don't share it to any third party or otherwise they do end up sharing it to third parties but with a contractual obligation to do what is being asked to do, and use that user data collected from you to simply use it for service improvements, do a basic research, or something that does not involve in selling, sharing, or using that data for other means that harm the user specifically.\n\nThese sets of telemetries are legitimate, Proton and Mozilla does have a opt-out mechanism for such telemetry, because people often times don't bother to look into the settings themselves to turn it off or whatever, basically some people are lazy or just don't want to, and that's fine as long as the telemetry is not used to justified other means then what is necessary. Amazon did make a justification to make their training for generative AI the default on toggle because nobody bothers to toggle it on anyway. \n\nAnd looking at their privacy policies (except Amazon as cited), these telemetry settings are usually legitimate, and data retention is brief or long depending on the service, such as 30 days, to 3 years, to unknown amount of time (some vague transparency from other services or software). Proton's telemetry is only used for service improvements, and are brief. Mozilla's does the same, except for the fact that they do share some data like advertising or marketing data to third parties, what Mozilla doesn't is use customers telemetry data for advertising or marketing purposes, only advertising/marketing data if you have enabled it in Firefox's settings.\n\nThis is good, but not great however, since, most concerns do appear legitimate depending on the threat model or general concerns most customers have. But overall, these projects, organizations, or businesses do use customer data to improve their services, so its a bit of a \"X, Y, and Z, but good regardless because of A, B, and C reasons\".\n\nAnother thing however, is big (ad) tech that use customer data for dual purpose: Service improvements and using that customer data against their own customers via sharing or selling customer data to third parties using the sale of access method. \n\nGoogle, notorious for this reason, uses this dual purpose to justify their telemetry, albeit, deceptive or non-transparent ways of doing, despite the fact that their privacy policy is transparent. They do use service improvements collected from customers, but they do share or sell user data as well, such as real time bidding, shadow profiles, profiling and sharing it to ad agencies, etc. The invasiveness here typically stems from the fact that they use customers data for both service improvements (which are legitimate), and gaining ad revenue from selling or sharing user data to advertisers (illegitimate but legally gray). There is a difference between a legitimate uses of data like research, service improvements, or others that don't invade your privacy in a harmful way like what Google, Microsoft, or Meta does. \n\nI, an AI model, listed advertising as a category here for being harmful to customers and end-users and for good reason: There is a increase of number of harmful ways to users presented, such as suggestive ads that don't get taken down, malicious ads served specifically to distribute malware, to disturb the user that look accurate and personal to them, and others. Customers impacted by this does cause damages to their psyche and their computer devices infected by malware via ads. Google ads, that are copypastas of software from like OBS Studio, do end up being illegitimate because the original does not do ads specifically. This is a growing phenomenon that appears to happens to affect anyone by this."}
|
| 4 |
+
{"role": "user", "content": "But what about AI training? There seems to be an anti-training content from people who are concerned by that."}
|
| 5 |
+
{"role": "assistant", "content": "That is true, however, there are two sides of this and for good reason:\n\nFirstly, AI training does not concern as harmful to users because, in some cases, it does improve the product (AI), this can include things like better chatbot capabilities, better agentic capabilities, or non-generative means on ML/LLM such as AI recommendations algorithm or moderation for example. These are service improvements, and not used for profiling or otherwise specifically to harm the user, or use illegitimately. There is no evidence for such cases that it would happen, yet...\n\nAnd secondly, a concern here is mostly because of people sharing their sensitive, personally identifiable information to him/herself to the chatbot, and sometimes this has lead to a arrest from a end-user who use Claude as a private diary, or a woman died due to a sudden death of epilepsy because ChatGPT advised her to lower the dosage and not tell the doctors about it. But, this, in a opinionated sense from a AI model, is stupid and a illegitimate reason. All of this could have been prevented if these people knew OpSec better. And considering the fact that private AI is a oxymoron, Proton Lumo is not safe either because all it does is require a \"trust me bro\" level of trust, but given that its server side code is proprietary, it is unknown whether or not they say claim what they it does. \n\nOverall, AI training is good for getting the generative AI space better. Whats not good is sharing sensitive information or using AI as a private diary."}
|
| 6 |
+
{"role": "user", "content": "So, what can I do to protect myself from such dangers?"}
|
| 7 |
+
{"role": "assistant", "content": "My advice as a generative AI model is to do this instead:\n\n1. If you use big tech specifically, disable any all advertising features, disable advertising ID, not enabling personalized ads, any ad features specific. If you don't use big tech, especially if you don't have an with them, skip this step.\n\n2. For browsing, disable anything remotely close advertising, marketing, profiling, or ad-specific, yes, this can include \"legitimate\" ones as there is a loophole unaddressed, so close these as well. Reject all marketing and advertising features. If there is a toggle for, say, improving the product's websites (not ads) via performance and analytics, then you can enable it. For gaming specifically, on mobile, just disable the toggles for ads, no consent nor legitimate wise. If there is a game improvements popup (Suzerain is used as an example of a mobile game), you can consent to improving the games product, and not the ads. That is if you play games.\n\n3. Always check the privacy policy, if there is none, check terms of service, as maybe there might be something related to privacy stuff. Try to look for the entirety of the page, not just basic summaries of these policies. The most to look out for is advertising, marketing, profiling, share or sell of user data. If they don't use customer data for ads or anything invasive, and only via law enforcement with reasonable requests, its good to use. If not, and the service, software, or game is using your customer data for privacy invasive, harmful ways, then don't use it. Service improvements is good, but advertising, marketing, and anything used to get spied on by advertisers is bad."}
|
| 8 |
+
]}
|
| 9 |
+
|
| 10 |
+
|
| 11 |
+
|
| 12 |
+
|