DaphneJan 28, 2023

RT @[email protected]

ChatGPT is great at its job. But its job is to be a bullshit generator -- an automated creator of plausible lies.

"It is very good at being persuasive, but it’s not trained to produce true statements" @[email protected] says in my latest newsletter. /1

https://themarkup.org/hello-world/2023/01/28/decoding-the-hype-about-ai

🐦🔗: https://twitter.com/JuliaAngwin/status/1619324373705560064

Decoding the Hype About AI – The Markup

A conversation with Arvind Narayanan

Show threadDaphneJan 28, 2023

The 1990s Internet drove the cost of *duplicating* content close to zero. Legal upheaval in *copyright* ensued.

Tech like ChatGPT drives the cost of *creating* slick, plausible content with no truth-value close to zero. What legal and cultural changes should we expect? 1/

Show threadDaphneJan 28, 2023
I assume plenty of people have written about this. I’m a bystander. But this seems to me like potentially good news for already-reputable sources. In a world of ubiquitous deepfakes, I’d pay to see which newsworthy images came from Reuters, or are vouched for by the WSJ. 3/
Show threadDaphneJan 28, 2023
I had a horrible moment where I thought this might mean we’d all rely on blockchain technology. @[email protected] talked me down. There are simpler authentication tools. (I’m not here to fight about this one. Take that somewhere else, pls.) 4/
Show threadDaphneJan 28, 2023
This seems like terrible news for the free exchange of information among strangers on the Internet. If I’m right about this, it is horribly sad. Will the prevalence of plausible fakes mean we never trust how-to videos on YouTube, or book reviews on Goodreads, even a little? 5/
Show threadDaphneJan 28, 2023
Do technologies like ChatGPT drive a world where trusted sources have new gatekeeper power over production of information that is valuable only if true?
Goodbye, helpful oddball on YouTube. Hello, professionally produced content from Good Housekeeping or Popular Mechanics. 6/
Show threadDaphneJan 28, 2023
What about first-hand accounts and witness videos? Police shooting videos? If we go back to a world where no allegation can be proven, and it’s just a he said/she said between police and people from marginalized or disempowered groups… Well, we know how that went last time. 7/
Show threadDaphne
Maybe that means: Upload your witness video immediately, to a “trusted” source like YouTube that will time/date/location stamp the upload. Subpoenas and warrants ensue. Surveillance —> authentication of truth claims.
Sigh.
This is not the kind of sci fi I want to read. 8/
Jan 28, 2023 at 3:29pmMastodon Twitter Crossposter
Show threadDaphneJan 28, 2023
Apologies to whoever has already written about these points. Including @[email protected], who was writing about evidentiary authentication of deepfakes half a decade ago. Like I said, I’m a bystander. This stuff is interesting/fun/scary/depressing. I hope y’all solve it. 9/9
Show threadDaphneJan 29, 2023

My 14 year old helpfully reminded me that there is nothing new in the world. Representative media has always been hell of fake.

https://youtu.be/gYGUfg_NJzg

Exposing Victorian Influencers Who 'Facetuned' Their Photos. (Photo Manipulation was EVERYWHERE 🤯)

YouTube
Show threadAaron WilliamsonJan 29, 2023
@daphnehk this is super fun, thanks for sharing!
Show threadJarred TaylorJan 29, 2023
@daphnehk me when I hear your son is 14 now
Show threadRobert ThauJan 28, 2023
@daphnehk Youtube can authenticate when a video was uploaded, but it can't authenticate the content of the video (might've been altered before upload). And uploading the video tags the Youtube user who did that -- which could be trouble for them if it depicts government abuse. And there's also the risk that they could be pressured by abusive governments to block or take down video evidence entirely, like so... https://www.businessinsider.com/youtube-twitter-blocking-bbc-documentary-indian-prime-minister-modi-2023-1
YouTube, Twitter blocking BBC documentary about Indian prime minister

A senior advisor to the Indian government called the BBC documentary "hostile propaganda" on Twitter and said videos and links sharing it are blocked.

Insider
Show threadAndrew AylettJan 28, 2023
@daphnehk
With ubiquitous network connectivity, I can imagine being able to securely attest that the first frame of a video was created after a timestamp and the last frame created before a timestamp.

That technically closes the door to any deepfakes which aren't created in real-time at the time of the event. I'm sure it wouldn't stop people from spreading misinformation though.