4 ms·

There is, depending on the type, format, and volume of data. Simply classify data into buckets for each value of each variable such that each unique combinatio

There is, depending on the type, format, and volume of data. Simply classify data into buckets for each value of each variable such that each unique combination of buckets for every variable contains at least N items for a suitably large number of N. It is then mathematically impossible to match an individual data point to any set of original data points less than N.

robin_reala 3y ago
I’ve read this three times now and I still don’t understand. Can you explain it in a different way?
- 65a 3y ago
  https://en.wikipedia.org/wiki/K-anonymity https://en.wikipedia.org/wiki/K-anonymity Basically keeping only chunks of text which have are not unique and contributed by many users.
  - bastawhiz 3y ago
    > Given person-specific field-structured data That's not what this data is and that's broadly not what LLMs are trained on.
    - 65a 3y ago
      By chunking and binning the data by user contribution count, you are structuring it as described. Pretraining is often accomplished on similar chunked text. k-anonymity is not perfect though, or state of the art, the wikipedia explains several known attacks.
tdudhhu 3y ago
[flagged]
pas 3y ago
it's still leaking information. let's say there are 5-6 companies that all work in the mystical cloud-painting sector, you lump them together and publish data about the sector. before there was nothing available now there's an average. and obviously partial information is very useful for exactly those with some other partial information. this is why OSINT is very effective. it's possible to piece together enough constraints on these datasets that in the end the likelihood of a successful deanonymization is unexpectedly high.
bastawhiz 3y ago
> Simply classify data into buckets for each value of each variable such that each unique combination of buckets for every variable contains at least N items for a suitably large number of N. What are the variables and what is the value of N when you're talking about "freeform text contracts containing PII"?
BobaFloutist 3y ago
What if I've been the plaintiff for thousands of labor disputes, and my name and SSN appears in legal documents thousands of times? Couldn't that potentially expose my name and SSN both 5213 times, demonstrating them to be related? I know this is a bit of an absurd example, but I'm trying to understand the mechanics more than to provide a realistic counterexample.

robin_reala 3y ago

I’ve read this three times now and I still don’t understand. Can you explain it in a different way?

65a 3y ago
https://en.wikipedia.org/wiki/K-anonymity https://en.wikipedia.org/wiki/K-anonymity Basically keeping only chunks of text which have are not unique and contributed by many users.
- bastawhiz 3y ago
  > Given person-specific field-structured data That's not what this data is and that's broadly not what LLMs are trained on.
  - 65a 3y ago
    By chunking and binning the data by user contribution count, you are structuring it as described. Pretraining is often accomplished on similar chunked text. k-anonymity is not perfect though, or state of the art, the wikipedia explains several known attacks.

tdudhhu 3y ago

[flagged]

pas 3y ago

it's still leaking information. let's say there are 5-6 companies that all work in the mystical cloud-painting sector, you lump them together and publish data about the sector. before there was nothing available now there's an average. and obviously partial information is very useful for exactly those with some other partial information. this is why OSINT is very effective. it's possible to piece together enough constraints on these datasets that in the end the likelihood of a successful deanonymization is unexpectedly high.

bastawhiz 3y ago

> Simply classify data into buckets for each value of each variable such that each unique combination of buckets for every variable contains at least N items for a suitably large number of N. What are the variables and what is the value of N when you're talking about "freeform text contracts containing PII"?

BobaFloutist 3y ago

What if I've been the plaintiff for thousands of labor disputes, and my name and SSN appears in legal documents thousands of times? Couldn't that potentially expose my name and SSN both 5213 times, demonstrating them to be related? I know this is a bit of an absurd example, but I'm trying to understand the mechanics more than to provide a realistic counterexample.