OpenAI’s hunger for data is coming back to bite it

https://www.technologyreview.com/2023/04/19/1071789/openais-hunger-for-data-is-coming-back-to-bite-it/

by tim_sw • 3 years ago

77 40 3 years ago

That thing with record labels going after people using generative AI on their IP caught my attention a little while ago. This seems like a big Achilles' heel for these products, but I guess the hope is reaching critical mass before anyone complains and then getting the rules changed in their favor, as a number of other tech products managed.

classified• 3 years ago

If you're big enough and rob enough people, your thefts will go unpunished.

pharmakom2• 3 years ago

OpenAI will have no trouble claiming legitimate interest. It’s very broad - broad to the point where HR companies are openly selling scraped LinkedIn data to recruiters.

gumballindie5• 3 years ago

Lets legalize theft because some get away with it.

dr_dshiv2• 3 years ago

Let’s redefine theft instead?

supermatt• 3 years ago

That’s simply not true.

Although legitimate interest does not specifically require consent it DOES require that the subject is informed of the collection of the data.

Also, legitimate interest doesn’t apply to many categories of data otherwise protected by law. e.g. health info

Also, it doesn’t override their other lawful rights, such as the right to be forgotten.

pharmakom2• 3 years ago

Although legitimate interest does not specifically require consent it DOES require that the subject is informed of the collection of the data.

Link? This was not my recollection of it at all.

Would also put Google and other search engines in violation so doesn’t seem right.

bonzini2• 3 years ago

Based on the injunction from the Italian privacy guard (the second one, not the ban) they did indeed back off on the scraping part, because it is not mentioned among the tasks that OpenAI has to do before April 30, for the ban to be lifted.

That said, this article is only devoting a sentence to way more important requests, which are the implementation of right to be forgotten and especially the problem with inventing false personal data. Which is a huge issue and what distinguishes artificial intelligence from autocomplete.

For more information: https://www.reuters.com/technology/italy-lift-curbs-chatgpt-...

By the way, blocking ChatGPT from Italy is not enough to avoid the GDPR violation. OpenAI is handling(*) personal data of Italian citizens, and therefore they must allow them to exercise their rights, even if OpenAI is not providing service in Italy. Blocking ChatGPT was just a fig leaf to show they were doing something.

(*) Because people say the personal data is not part of the model, "handling" is defined as performing any in a list of actions which includes disseminating, and "personal data" is defined as "data that allows identification of a person". So if "ChatGPT is disseminating data that allows identification of a person", then "OpenAI is handling personal data"; the former is simply a subset of the latter.

cfn• 3 years ago

I really hope we don't get a new version of the "You cannot view this webpage because you are in the EU", this time for LLMs.

bonzini2• 3 years ago

That would not be enough. You need to implement the right to be forgotten of European citizens even if you are not offering services in the EU.

Rather, I wonder what would be the effect of this on open source weights, as they have the same issue of being able to produce personal data of European citizens.

charcircuit• 3 years ago

even if you are not offering services in the EU.

If a company isn't operating in the EU then there is little the EU can do to them as the EU doesn't have jurisdiction over them.

cozzyd• 3 years ago

As an Illinois resident, I'm going to guess I'll be getting a check from openai for some biometric privacy violation at some point in the future...

bottom999mottob• 3 years ago

They've already got red light cameras all over Illinois, so I wouldn't be surprised

MiyamotoAkira• 3 years ago

GDPR is the best thing since sliced bread

vitehozonage2• 3 years ago

Good. I wish they would be fined into bankruptcy

desertraven• 3 years ago

Just curious, why?

vitehozonage2• 3 years ago

Even my own content has probably been used by them, against my wishes. I see it as unethical data usage. Additionally, the ethics of the company's history, changing to become entirely 'closed' while even keeping the name "OpenAI", makes me... angry, I suppose is the right word. Intuitively it seems to me that they will have a very bad influence on the world if they maintain dominance and continue to gain power

MrBuddyCasino2• 3 years ago

I'm glad to see we're getting better and better at regulation. Here we have the most important development in computing in years, and we're immediately nipping it in the bud. There used to be some delay.

soco2• 3 years ago

Same with the atom bombs, we only got to drop them twice, what a beautiful show we missed!

MrBuddyCasino2• 3 years ago

The analogy you’re looking for is the broader term Nuclear Power, and yes same thing except we allowed it to happen for a few decades before switching to burning coal instead, thanks for the reminder.

meepmorp• 3 years ago

This, but unironically. We should’ve built more and done Operation Unthinkable.

corbulo• 3 years ago

Are large datasets too much exposure or worth the risk? For OpenAI it appears the latter.

svaha1728• 3 years ago

I cannot verify, but delving deep into electrical engineering topics I’m guessing most of that training set is coming from expensive college text books and papers that are behind paywalls. My guess is it scraped all of Sci-Hub. Which is a great resource, albeit illegal.

OpenAI’s hunger for data is coming back to bite it

Related Stories