May 5, 2024

The mountain of shit theory

Uriel Fanelli's blog in English

Fediverse

Guarantor up, Guarantor down.

Since it is said in the Italian newspapers that the Guarantor has closed ChatGPT to the Italian public, I feel compelled to clarify the situation a little. Because what happened is anything but "the guarantor who does things". First, we must try to understand what a "guarantor" or an "authority" is in Italy.

In theory, the law in Italy does not follow the common law. This means that if we forbid something, and we say it's a criminal violation, the judge has to proceed ex officio. And he doesn't have to proceed as he pleases, because there is a code of procedure that must be followed.

Well.

But Italy is the country it is. It means that the current politician wants "guy A" to be punished, but "guy B" not to be punished. In short, we want the law to be applied to enemies and interpreted for friends.

The way the system of division of powers is built, the thing is difficult, or at least problematic. Certainly risky.

So: if I want to allow the call center mafia to call you at home despite the law, but I don't want a foreigner to enter the market, what do I do? Do I bribe a judge? Do I bribe all the judges?

No. I act as a “Guarantor”. An 'Authority', an 'authority'. Why'? Why

  1. From that moment the Authority decides, in an arbitrary manner, whether and against whom to proceed or not.
  2. The authority is appointed by the politicians, and answers to them.

This way, it's possible to have BOTH the call center mafia running rampant, AND the "Guarantor" beating up ChatGPT for "the data", without specifying WHICH data, since no one knows what data he's talking about.


Let's take a concrete case. Privacy law in Italy is law. So the judiciary, which HAS the power (they can even put you in jail, or fine you), could in theory take care of the thing by itself.

So, if they fuck you up with harassing phone calls from call centers, you could (if you feel like doing the favor) report the matter with a lawsuit, and/or go to a lawyer and ask for damages, as they may make your phone unusable. work cell phone.

And the judge, if he proves you right, could sanction them. Or, the magistrate who investigates through the Postal Police could easily trace the origin of this business, and discover that the Italian telco companies make a lot of good money from it.

Instead, you have to save the ass of the friends of the telcos who sell the numbers and access to the HLRs with which they can make anonymous calls. Thus a "Guarantor" is appointed who never proceeds, never investigates, and even if he did he would have no power.

And the consequence is that while the guarantor takes care of ChatGPT, your phone is bombarded. If ChatGPT were a Call Center, the "Guarantor" would probably look the other way.


Having said what a "Guarantor" is in Italy, let's see what happened.

What did the Guarantor actually do? First, it is NOT about the data you use to train the language model. For two reasons:

  • The Privacy Guarantor does NOT deal with copyright. Whether or not ChatGPT had the right to use Wikipedia as training data is not its problem. At most of SIAE.
  • From this training data a Language Model is extracted, or if you prefer a BIG Language Model. Which, not containing PII, is certainly an AGGREGATE data, which consequently escapes the meshes of the GDPR, which leaves great freedom on aggregate data.

Example:

  1. "Giovanni the butcher of Via Biella 12 is gay" contains both PII and highly protected data, so the GDPR has something to do with it.
  2. "In our country 3% of butchers are gay", even if the country had 33 butchers, it does not contain PII, and the figure is an aggregate.

In the second case, the GDPR allows the use of the data.

Since the linguistic model synthesizes enormous quantities of data in an incredible way, it is certainly an aggregate, therefore the Privacy Guarantor has nothing to do with it.

But there's another problem. And here the Guarantor has something to do with it.

  1. An account is required to interact with ChatGPT, and OpenAI has never shown “Privacy by Design” documents that are required.
  2. When you are inside, you chat, and those chats are kept (so much so that you will find them again the next time you log in).

Wikipedia's intellectual property has nothing to do with it, because the Guarantor doesn't deal with it.


And here you understand that there is a problem here. On the one hand there is your login, with which you are identified, therefore it is a PII. On the other hand, you asked ChatGPT if it's normal to shove a pineapple up your ass before watching Inspector Derrick, and you also said that you do it every Friday. And this is a data protected by the GDPR.

But this does not give the Guarantor the power to ban something: in Italy the guarantors must not have power , precisely because they are made to REMOVE power from the law. However, they can "ask for clarifications", an operation that can break anyone's shit, for months and months, and slam the offender on the pages of the newspapers.

The Supervisor, therefore, worried about your ass, about the pineapple and also about Inspector Derrick, simply asked for clarifications under the GDPR. Of course, he expected the other side to say:

"but of course, useless redundant and ineffective pseudo-institution of a nation that doesn't count for shit on a technological level: we can't wait to waste precious time during the most important and crucial technological race of the new millennium to protect the Italian ass from the evil pineapple , and we yearn to be able to waste time and resources with your petty bureaucrat bullshit from overseas.

Instead, in OpenAI the debate was this (the spoken language is a mix of American and Indian slang):

  • Hey Boss, we received a letter from the “Guarantee of della Praivaciai of Italy”. (literally)
  • HEY GUYS! Who ordered all the pizza for lunch? Is here, Hurry Up!
  • No, boss. It seems more kinda legal stuff. Lot of papers, not pizza.
  • What? paper? In 2023? Who the fuck is still using paper in 2023?
  • Well, seems they do. It looks like they mention some legal obligation we are supposed to fulfill.
  • We? Why in the name of God we have legal compliance in Italy? Fuck, fuck, fuck, I told you: go sushi, Damn Sushi! Clean, straight sushi! No legal obligations, no strings attached. But no, all that weirdo wants pizza!! Pizza makes you fat! And now we have this! Because of Damn Pizza!
  • It's not about pizza, boss. According with some weird english I read, it seems they want us to grant some privacy to people who likes to stick pineapples up their bums.
  • Wa… what?
  • Yes. Apparently, they want to be sure that people chatting with ChatGPT about sticking pineapples up their bums will be granted some privacy.
  • Please tell operations to remove ANYTHING about “pineapples up the bum” in training data, then train again, then answer that we are clean. Done.
  • Er… may not suffice, boss. Seems they take very seriously the issue of pineapples up the bum.
  • Okay. Don't panic. Tell me how much traffic from Italy we have.
  • Apparently about 0.0003% of traffic.
  • And how many revenues we do in Italy?
  • No revenues at all. They have no money, they are poor.
  • Okay. I want Italians out of my yard, block the access to the system to the frigging country, no more Pizza at lunch in this company, and then let's go to our Weekly Office Yoga meeting. We have more important things to discuss.

In short, they're in the biggest tech race of the millennium, and they don't have time for Supervisor bullshit. In terms of business, Italy is not relevant, so they limit themselves to removing the tooth and blocking Italy.

Result: OpenAI has more important things to think about than talking to the Italian "Privacy Guarantor", and prefers to block Italy (quick and simple) rather than enter into a legal dispute on the size of pineapples.

As for pineapples, you can always write to Cioe' if you have any doubts. I can tell you that you can't get pregnant by shoving one up your ass. Not even if your boyfriend used it before.


Let me be clear, they could also have been tougher, and eliminated Italian from the languages ​​in use by Chaptgpt, putting an "end" to the free use of the product by Italians. And let's be even clearer, it's not even certain that they won't do it in the future. And here Italy would really be behind in a crucial technological race. (for a change).

But the thing that irritates me the most is listening to podcasts where "the algorithmic is political", in which the usual politician tries to say that the problem of the Privacy guarantor would be the data used to train ChatGPT, even there enshrines the right or not to use Wikipedia, forgetting that the PRivacy Guarantor does NOT deal with licenses and copyright at all.

Technology is NEVER politics. And it's not because it first of all requires competence.

While politicians, well, they really don't believe in competence.

Leave a Reply

Your email address will not be published. Required fields are marked *