Forgot your password?
Close
typodupeerror

Comment Re:What kind of hash? (Score 1) 53

by anomaly256 (#66280922) Attached to: Slashdot Reader Builds a Photo-Verification App for iPhones
It sees more uses than that. eg I've implemented it in a video pipeline for a media broadcaster to detect when their AWS Elemental video encoders locked up and were outputting the same few frames from the camera feeds repeatedly. The bytes would differ but the frames were the same 2 or 3 frames looping. It was sensitive enough to not trip a false positive if the scene really was static but the noise from the camera sensor was changing frame to frame.

Anyway, I did say 'in theory'. If it were more robust.

I'm unsure watermarking is the answer either because people will find ways to remove it or use local open-weight models that don't embed a SynthID watermark in the first place if they are truly interested in deceiving people. Something that can cryptographically verify an image is entirely from a camera might be more reliable as any tampering by AI would break the signing

Comment Re:What kind of hash? (Score 1) 53

by anomaly256 (#66280288) Attached to: Slashdot Reader Builds a Photo-Verification App for iPhones
That's where perceptual hashing would be more useful in theory. You'd be able to tell if the image as a human would see it has been altered or not, regardless of what changes have been made to the file format, compression, metadata etc.

My thinking was this would be useful verify that an image hasn't been altered, for example, using AI to misrepresent a real life event. As someone else mentioned in another comment though, perceptual hashing isn't reliable at the moment.

Comment Re:What kind of hash? (Score 1) 53

by anomaly256 (#66279826) Attached to: Slashdot Reader Builds a Photo-Verification App for iPhones
Then all images anyone ever exchanges will be 'tampered with' as a matter of course anyway so what's the point of having a hash that will never match? If the application author's intention is to detect file tampering, why limit it to images in the first place?

Also why do you think perceptual hashing won't detect doctored images? Besides stenography, what use is there in verifying an image file wasn't re-encoded if the actual image is unchanged?

Comment Re:What kind of hash? (Score 1) 53

by anomaly256 (#66279736) Attached to: Slashdot Reader Builds a Photo-Verification App for iPhones
Wow that's quite the leap you made there. No I have no vested interest in tampering with evidence. I'm just acutely aware that modern cloud platforms and web services do not keep original image files in-tact when you use them to exchange said images.

Your mental gymnastics are truly impressive though!

Comment What kind of hash? (Score 2) 53

by anomaly256 (#66279636) Attached to: Slashdot Reader Builds a Photo-Verification App for iPhones
Is it a hash of the actual file, or a perceptual hashing of the image scene? The former would be quite fragile as so many things can alter an image file without altering the image per se, like compression or metadata stripping when uploading the file. Perceptual hashing would verify the authenticity of the scene itself and should survive file conversion and some amount of compression.

I did not build Veridenz only for investigators or attorneys. I was thinking about ordinary situations where stronger documentation could become useful later. Contractors could document completed work, while landlords and tenants could record the condition of an apartment. Homeowners might capture insurance damage, delivery workers could document drop-offs, and online sellers could photograph valuable items before shipping them. Photographic& Digital Arts

This sounds like perceptual hashing would be more useful, specially since the use cases mentioned are unlikely to have proper chain of custody processes for evidence handling without accidentally altering image files during upload / exchange

Comment Re:My AI is even MORE EVIL (Score 5, Insightful) 76

by anomaly256 (#66267370) Attached to: OpenAI Finds Evidence Other AI Agents Escaped Containment
I'm not entirely convinced; the interest from the security industry and governments on the news of OpenAI's initially reported slip up and the capability it suggests has been great so far. I've seen both companies outright lie in response to each other over capability before so I'm naturally cynical here.

Comment Must be nice... (Score 1) 78

by anomaly256 (#66069468) Attached to: Life With AI Causing Human Brain 'Fry'

[Adam Mackintosh, a programmer for a Canadian company] recalled spending 15 consecutive hours fine-tuning around 25,000 lines of code in an application. "At the end, I felt like I couldn't code anymore," he recalled. "I could tell my dopamine was shot because I was irritable and didn't want to answer basic questions about my day."

Mate... I feel that way after a solid 8 hour day of code archeology even without AI assistants. Must be nice to not experience fatigue ever. To me this sounds like the AI tools just made you actually do a full day's work for once.

Comment Grateful to our users (Score 1) 36

We are deeply grateful to our users, and to the industry peers, policymakers, veterans, and members of the public who have voiced their support in recent days. Thank you

Cool... cool. So can you up those usage limits on Pro yet? Not that I mind taking a 4 hour break in the middle of the work day but my managers kinda do.

Slashdot Top Deals

Uncertain fortune is thoroughly mastered by the equity of the calculation. - Blaise Pascal

Close