This is unrelated, but http://theartbot.com looks great! Do you want to submit it to HN sometime? Something like "Show HN: TheArtBot - Semiautonomous drawing machine" would be a great title. It would also be great to include text that gives the backstory of how you came to work on this, and explaining what's different about it (e.g. going into some of the details that you mentioned at https://news.ycombinator.com/item?id=27229274).
> an adversary could trick Apple’s algorithm into erroneously matching an existing image
This is a very real, possible attack. Apple ships its CSAM model on device so any attacker can have a copy of the model. Then the attacker creates an image that triggers CSAM but looks like a panda [1]. Now the attacker sends tons of triggering photos to the unsuspecting victim, who now gets questioned by the FBI.
> Now the attacker sends tons of triggering photos to the unsuspecting victim, who now gets questioned by the FBI.
That's glossing over the middle part where a human from Apple (before it even gets to law enforcement) actually look at the images and goes "oh, these are actually pandas" and realizes they were erroneously detected.
So the attacker creates an image then the user has to download it. Then the FBI digs in and see it was a crafted false positive, then begin to investigate who sent it and why. Then the user takes civil action against the person who sent it for harassment.
More precisely, 30 carefully crafted false positives. All of which need to be imported into your iPhone's photo library to sit alongside pictures of your dog and your mum. And then they have to get past human review. Not impossible, but so far beyond implausible that it can be dismissed as ridiculous.
And if this trick ever works, it could only be done once before Apple has the opportunity to plug holes in their NeuralHash algorithm and fix any deficiencies in the manual review process.
I wonder if shops will still install used (stolen) units regardless? Or maybe they are just being shipped out of state or broken down for raw materials.
C'est la vie. You're not supposed to steal cats either.
The one thing that's fairly hard to get around is the biannual inspection.
In days of yore, a box of donuts could get someone to hot pipe your car. Now, not so much. It was kind of a necessity if you owned a car that wouldn't pass even in absolutely pristine new condition.
Are you asking what is the advantage of having ML optimized hardware on device? Yes, running inferences is expensive too, especially for video, photo, and speech processing. I would expect this phone to have user noticeable improvements in those three areas.
> The average cost of housing has gone up over time, but this trend is mainly driven by the fact that modern houses are larger and built better than houses in the past.
Oh, that explains why the house that sold for $300k in 1990 sold today for $2M (only 3X adjusted for inflation). It must have gotten 3X larger and 3X better built, even though it is the exact same house with a couple extra layers of paint and a redone kitchen.
Nice! I did something like this a long time ago: http://theartbot.com
I wrote a couple algorithms but the best one basically treated the darkness of the image like the altitude in a map, then drew contour lines. I also physically built the plotter.
Likely because Google Assistant did not know it was running on a tv. It's probably the generic assistant that comes with Android. Not sure I would consider that pathetic. Tuning all the intricacies of the Assistant for each application seems like a ton of work.
I love these projects.
I did a related project where I created a single, solid object that has 4 different silhouettes/shadows from 4 different angles in the same plane. Using more view angles than dimensions makes the problem much more complex. I suspect it is not possible to do this purely programmatically so I did it by hand had to smudge the letters to get everything to work out.
Thanks for sharing. Yes, what we found is that multiple input images often are not consistent with each other (i.e., no physical volume that can satisfy the constraints from all the views). We thus start with a set of consistent voxels and connect them so that the projections are as similar to the inputs as possible.
ZKM in Karlsruhe, Germany has an exhibition called Open Codes, where they have used a single solid object to generate all letters in the alphabet using projections.
There supposed to be a spiral encoded in the first chapter of genesis. And the spiral itself encodes the alphabet the encoding of the spiral is written in. Very quine like ability. I am not familiar with any other alphabet achieving this.
http://meru.org/letteressays/letterindex.html
Please, it doesn’t help anything to publish unfounded conspiracy theories. Apple already publishes open source projects and has a bug bounty in which they pay for security vulnerabilities in some areas of iOS. It’s ludicrous to advance the theory that they hatched a plan to leak this important project to Reddit instead of through one of their official channels.
Thank you so much for the kind words. I am the creator and love these kinds of projects. Check out another project I created along the same lines: TheArtBot.com
If you are asking why not many people do these types of projects- you sounds like an excellent candidate to do one yourself! I have list of project ideas on my phone that I will add to at random times. Next time you think of a cool idea write it down and make it a reality!
According to my page, you called it "Super Cool".