You could, for orders of magnitude more compute than reading a magnetic encoder (my assumption at how they estimate it)

Sure, but not more than what you're already spending on transforming the image. And it's not like these devices are exactly lacking in horsepower.

This is trivially broken by people who affix some type of cover over the camera. I do this on the off chance some errant application thinks it deserves to take pictures of my environment.

If someone covers the camera, the feature isn't relevant since it requires the camera to see your desk

Isn't the desktop view is produced from the iPhone camera capture, not from from the MacBook's camera?

If you have a new Macbook the built-in camera does it

I'm typing on a 2024 Macbook Pro. Is that sufficiently new? I don't see how it would work, practically. The only camera is the user-facing one. If the screen were tilted down toward the desk, I'd have to kneel down to see it.

But compute is cheaper for the manufacturer than adding a sensor (parts & labor, and it adds up over millions). Someone must've done the math.

The Mac camera light is wired inline. If the camera is on, so is the light. Since we're not seeing the camera light flashing on periodically, this isn't how it's being done.

The Macbook tally light isn‘t necessarily wired to the camera. It very well could be independently software controlled. At least it was not too long ago. IIRC there was an article about this, posted here on HN.

Macs used to have (still have?) a feature where you could declare it as lost/stolen and remotely take a photo with the camera. I believe the light didn‘t glow for that.

shameless plug: https://sannysanoff.github.io/whiteboard/

not only for mac users.

Relevant XKCD: https://xkcd.com/1425/

This was correct a number of years ago. Feels a little strange we can just do an API call for bird recognition now.

But is there actually an API for that? Last I checked the big providers Video Intelligence APIs even distinguishing cats and dogs was still unreliable.

Just to see if a bird is in the picture (like the comic states) using chatgpt et al can probably do a sufficient job.

Not condoning people make this app, just thinking about how fast things have moved in just a few short years.

For a POC, I’ve done animal recognition in a picture with Anthropic and the various Amazon Nova models. It’s around 10 lines of code.

BirdNET from the Cornell lab of ornithology provides that api.

Unless I am missing something massive, BirdNET[0] is for identifying birds by sound, not by images.

Merlin[1] (also from Cornell Lab of Ornithology), on the other hand, has both image and sound ID. I haven't used either, so I cannot compare the quality of results from Merlin vs. BirdNET for sound ID, but afaik only Merlin has image ID.

0. https://birdnet.cornell.edu/

1. https://merlin.allaboutbirds.org/

These days you dont need an api, you can run the stack on tamagochi

Flickr did it in 2014, same year as the comic. Unfortunately the service is down and they didn't include a screenshot of it working.

https://code.flickr.net/2014/10/20/introducing-flickr-park-o...

Ho boy, good luck convincing people it wasn't watching them wank!

That sounds like an excuse to enable turning on the camera without turning on the light for it just because no user-software is using it. No thanks.

Plenty of users put stickers on their cameras. One simple user trick would break your whole workflow.

The Mac camera light is wired inline so as to make this impossible. The only way for the camera to be on and the light not is if the light itself is broken.