Tilmaaminta SynthID ee Google: Isku-dhac weyn oo ku saabsan xaqiiqada iyo khaldanka AI
Google ayaa sheegtay in aaladaha ay leedahay ay sameeyeen 100 bilion oo sawirro iyo muuqaallo AI ah, waxaana la isticmaalay tilmaaminta SynthID si loo xaqiijiyo xaqiiqada.
Sawirrada AI: Isbedel degdeg ah oo ku yimid taariikhda
Warbixinta Starling Lab, oo ka socota Jaamacadda Stanford iyo Jaamacadda USC, waxay sheegtay in qofka aad u qaadayo 1.5 bilion oo sawirro uu u baahan yahay 149 sano ka dib markii la sameeyay kamarada.
Laakiin, AI-ga generative wuxuu sameeyay 1.5 bilion oo sawirro kaliya 18 bilood gudahood. Tani waxay muujisaa isbedel degdeg ah oo ku yimid taariikhda sawirrada.
Tilmaaminta SynthID iyo C2PA: Laba hab oo kala duwan
Google waxay isticmaashaa labo hab oo kala duwan: SynthID, oo ah tilmaamo aan la arkin, iyo C2PA, oo ah nidaam metadata ah. SynthID waxay ku jirtaa pixelada sawirka ama muuqaalka, waxaana la sheegay inay noqon karto mid adag oo aan la burburin karin.
C2PA waa mid amniga ah, laakiin waxaa laga saari karaa fudud marka la sameeyo sawir ama la qaado screenshot. Google waxay sheegtay in SynthID ay noqon karto mid adag oo aan la burburin karin, isagoo ku jira meme-yada.
Xaqiiqada SynthID: Waxay u baahan tahay in la tijaabiyo
Pushmeet Kohli, oo ka tirsan Google DeepMind, waxay sheegtay in kooxda ay ku bixisay waqti badan si ay u hubiso in SynthID ay noqoto mid adag oo aan la burburin karin.
Laakiin, waxaa jira shaki ah in SynthID ay u baahan tahay in la tijaabiyo, gaar ahaan marka OpenAI, Runway, iyo Nvidia ay bilaabayaan isticmaalka SynthID.
Google has unveiled a durable invisible watermark to label AI-generated content, yet researchers caution that no single tool can contain the flood of synthetic media already being produced at unprecedented scale.
The Scale of Synthetic Media Is Unprecedented
It took humanity 149 years after the camera's invention to produce 1.5 billion images, but generative AI accomplished the same feat in just 18 months, according to Starling Lab, a research collaboration from Stanford University and the University of Southern California.
This spring, Google announced at I/O that its tools alone had generated more than 100 billion AI images and videos in a couple of years, a figure that underscores the explosive growth of synthetic content across the internet.
Two Approaches to Labeling AI Content
Google employs both SynthID, an invisible watermark encoded in the pixels of images or the waveform of audio, and C2PA, a cryptographically secure metadata standard that cannot be forged but is trivially stripped by editing or screenshotting.
Google DeepMind scientist Pushmeet Kohli told Ars Technica that the team invested significant effort into making SynthID durable enough to survive compression, resizing, and even circulation as well-worn memes.
Can Watermarks Solve the AI Content Problem?
As OpenAI, Runway, Nvidia and other companies begin adopting SynthID, experts question whether invisible watermarks alone can address the broader challenge of identifying and countering AI-generated misinformation.
The technology may offer a useful layer of provenance, but the sheer volume of synthetic media and the ease with which metadata can be removed suggest that no single solution will be sufficient.
Ilaha iyo xuquuqda sawirka
Sawir: Ars Technica Xigasho
