Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Does GPT-2 Know Your Phone Number?

bair.berkeley.edu

Post reply on HN
Print view View original

101–110 of 161 posts

Re: Does GPT-2 Know Your Phone Number?

#101

Post by acdha »

m463 wrote: post #71
dmm wrote: post #8

Earlier quoted context omitted.

So in terms of copyright, is GPT-2 a derived work of that recipe? Or generally are models derived works of their training data? It seems lots of people use training data from Flickr, like COCO, and then use the resulting model for commercial services.

Ethical companies make sure they have rights to the training data before making models. That said, now a lot of privacy policies make sense. ALSO, it makes me wonder about "Your call is recorded for training purposes"... is it a coincidence or is it very carefully worded?

C) Unchanged for decades - that wording predates speech recognition or neural networks being more than research curiosities. The big surprise was how quickly the possibility of a double meaning became possible.

Re: Does GPT-2 Know Your Phone Number?

#102

Post by courtf »

agumonkey wrote: post #99
courtf wrote: post #96

Earlier quoted context omitted.

It's one of the biggest lies in tech right now, that all this user data is some untapped gold mine, as if text on the internet is the most important measure of human behavior. It's all going to end up rotting on backups while the marketing department spins yarns about the amazing AI that's going to spring forth any day now from a pile of tweets. It's like trying to build a dog by hanging around the park collecting tu…

is the data market a bubble ?

Depends what you're doing with it, but I have to think it mostly is a bubble. Maybe if you have a really simple question that already fits what can be measured, or you're trying to surveil someone in particular, but for stuff like AI just doesn't make sense to me.

Re: Does GPT-2 Know Your Phone Number?

#104

Post by ben_w »

ClumsyPilot wrote: post #66
ben_w wrote: post #49

Earlier quoted context omitted.

If you make some fairly plausible assumptions about Moores law (and equivalents), the simulations are cheaper than the meat bodies before mid-century. • https://worldbuilding.stackexchange.com/questions/51746/when... • https://kitsunesoftware.wordpress.com/2018/10/01/pocket-brai...

That is cool, but tangential to the question posed above - given that every part of surveiliance capitalism is currently working against my interest, either to sell me something I dont need, to inflienxe the way I vote, etc. Will the simulation of my brain belong to a megacorp, and be used to do even more damage?

> Will the simulation of my brain belong to a megacorp, and be used to do even more damage?

Potential damage: the sim can do any work which meat-you might otherwise be paid for, building the things the megacorps want to build, while meat-you sufferers from .

Even worse outcome: the sims are sentient and they know they’re enslaved and they can’t do anything about it.

(I don’t think anyone yet knows if sims would be sentient, or if it might be optional, but in the worst case dystopian nightmare the combination would definitely be even worse than just one of the two).

Re: Does GPT-2 Know Your Phone Number?

#105

Post by Triv888 »

Zenst wrote: post #61

Friend asked me if Google knew his phone number and details, I said they do at a quantum level. If you don't look they may or may not have your details, yet if you look - you are giving the details to search for and then they would have them if they did not.

That's why when you visit rotten sites like mylife.com, the first thing they ask you is that if you are searching about you so that they can extract as much info. from you as possible.

Re: Does GPT-2 Know Your Phone Number?

#106

Post by anticensor »

jcims wrote: post #7

I was in a discord with someone that had direct access to GPT-3. We played Jeopardy with it, where we would take a wikipedia article about a person and use a snippet of the first paragraph as the prompt and ask GPT-3 who it was. It was very good at guessing the right person. I'm curious if one can use GPT-3 to do some clustering of people based on their writing. For example, if I take the writing of several diagnosed…

GPT is generalisable to other kinds of data. See ImageGPT for example.

Re: Does GPT-2 Know Your Phone Number?

#107

Post by geofft »

fouronnes3 wrote: post #4
newman8r wrote: post #3

Playing with the AI dungeon a while back (on the GPT-2 mode) I was presented with a tilapia recipe, titled "Kittencal's Broiled Tilapia" - it sounded bizarre so I decided to do a google search and I found that it was directly pulled from from https://www.recipezazz.com/recipe/broiled-parmesan-tilapia-7... - the user who posted it was 'Kittencal'

There is enough GPT discussion online now that the next language model will be trained on text that talks about itself. That's when it gets interesting.

This seems like what xkcd called "citogenesis" - something inaccurate gets published on Wikipedia, a traditional "reliable source" (newspaper, published nonfiction book, etc.) retells that bit of information without a source, Wikipedia cites the source, that information is now firmly established as truth for the rest of humanity.

I worry about this with Google's language translation models, too. It's entirely possible that it's making up phrases or connotations that never existed in organic human speech, but people who aren't fully fluent in the language use Google Translate for assistance, publish something, and then suddenly it's in a published text by an actual human and Google reinforces its own belief.

For at least the past five years, Google Translate has translated "who proceeds from the father" into Latin as "qui ex patre filioque procedit" - inserting the additional word "filioque," which means "and the son." The question of whether to add this word is a 1500-year-old theological argument: https://en.wikipedia.org/wiki/Filioque Since the Western Church added the word, most texts in Latin include it, so Google is almost certainly deciding that the phrasing with "filioque" is more popular - but it doesn't know what the words mean, so it can't realize that the phrase it came up with means something different!

Re: Does GPT-2 Know Your Phone Number?

#108

Post by emmelaich »

st1x7 wrote: post #91
visarga wrote: post #41

Earlier quoted context omitted.

Maybe this data will be used to 'upload' you in a virtual world after the singularity. It might be your second life.

Even in the science fiction version of this, we're still talking about a copy. It's not you, you can't leave your body or be uploaded. It's just a digital copy that looks like you.

You can destroy the original.

This would make a good SF story. Probably has already :-)

Re: Does GPT-2 Know Your Phone Number?

#109

Post by gwern »

px43 wrote: post #19

I've been saying this for years, but every post you make online, every unencrypted email, IM, text message, etc, will eventually end up getting sold off as training data for future machine learning projects. Every company stores this stuff for ages, and the value of candid conversation data just keeps increasing. Eventually these companies are either going to get hacked, get bought, or go bankrupt, and all the cleart…

The value of data decreases. There are diminishing returns to dataset sizes (specifically, for GPT, it's a power law). The more public data there is, and the more data in general, the less any additional piece of private data is worth. Why would any tech company with access to lots of private data, like Google, risk the enormous backlash and legal penalties when they can so easily scrape terabytes and terabytes of text from public datasets to train uncannily intelligent models like GPT-3? Even hobbyists can make the necessary datasets from Common Crawl et al (see EleutherAI's terabyte+ of clean text in The Pile dataset).

This is in addition to the ever greater sample-efficiency of bigger models, which learn eerily fast from just a handful of datapoints, rendering the supposed 'moats' of big giant proprietary datasets ever more moot...

Re: Does GPT-2 Know Your Phone Number?

#110

Post by xapata »

geofft wrote: post #107
fouronnes3 wrote: post #4

Earlier quoted context omitted.

There is enough GPT discussion online now that the next language model will be trained on text that talks about itself. That's when it gets interesting.

This seems like what xkcd called "citogenesis" - something inaccurate gets published on Wikipedia, a traditional "reliable source" (newspaper, published nonfiction book, etc.) retells that bit of information without a source, Wikipedia cites the source, that information is now firmly established as truth for the rest of humanity. I worry about this with Google's language translation models, too. It's entirely possibl…

I met an artist whose body of work consisted of citogenisis. He'd been at it for a decade when I met him, and had created at least a dozen fictional artists, detailed biographies, and a complete "retrospective" gallery show for each. This let him copy the styles of the great masters of different periods, yet make something new, of a sort. He never edited Wikipedia himself, but instead led gallery tours of these retrospectives and was always overjoyed when people would write about them, propagating the fiction.
Post reply on HN