why you should read my effective privacy tips and not use AI

Don’t go to ChatGPT or some other LLM-based service for privacy advice. Here are four reasons you should read my article instead. (or find somebody’s. I have links.) Whatever the LLM puts out is based on the documents in its training set, and the Internet has a lot of material that looks like credible privacy advice but is either expired or wasn’t really useful in the first place. Some of the problem material that will have been fed into an LLM includes:

1. Out of date information. Today, two big companies dominate the surveillance advertising industry and old advice about cookies and data brokers looks like advice about protecting yourself from junk faxes or something.

2. Privacy policies. Corporate privacy policies are full of ineffective advice, like industry opt-outs that, at best, worked back in the third-party cookie days (and often had errors even then).

3. Content marketing. For example, most of the advice about using a VPN for privacy applies to specific areas where a VPN is useful, like getting to sites that require age registration in some places or running BitTorrent. But the VPN content marketing people don’t want to mention pr0n or BitTorrent too obviously, so they write like they’re giving general privacy advice, and a lot of that material gets snarfed into training sets.

4. Sanitized, safe privacy advice that’s not practical. For example, it’s very common to tell people to be sure to read privacy policies, which nobody has the time to do and still participate in the actual economy. And the FBI recommends an ad blocking extension instead of recommending a known safe one, even though the ad blocking scene is full of malware, spyware, and, yes, adware.

A human-written privacy article will be able to avoid all that stuff. Mine has a list of tips that are easy to do, and, just in case, links to others that have different points of view. I also try to avoid writing like a bot. Here it is: effective privacy tips

Related

Apple vs. Facebook is Kayfabe by Alex Russell. (Don’t rely on tips that depend on choosing a privacy-friendly brand. Almost everybody has something. Much content marketing copy has been devoted to positioning some brands as privacy brands, but the reality can be different.)

LLMs and reputation management Why the AI summary of a biographical or historical subject is likely to have similar problems.

Bonus links

Did Anthropic AI steal your book? You can sign up for updates on the class action suit. by James Folta. As it winds it way through the courts, the lawyers representing the plaintiffs want to hear from authors who may be affected. Lieff Cabraser Heimann & Bernstein have set up a site where you can enter in your name, your books, and your contact info to stay in touch. Related: Authors celebrate “historic” settlement coming soon in Anthropic class action by Ashley Belanger.

My Case Against Meta Moves Forward by Kelly Stonelake. We’re moving forward! This means the case now moves into discovery, where Meta will have to turn over documents and testimony that could reveal what really happens behind closed doors.

We Are Still Unable to Secure LLMs from Malicious Inputs by Bruce Schneier. Any AI that is working in an adversarial environment—and by this I mean that it may encounter untrusted training data or input—is vulnerable to prompt injection. It’s an existential problem that, near as I can tell, most people developing these technologies are just pretending isn’t there.