Psync's Genie S security camera uses GPT to describe what it sees
If you ask Psync Labs, it'll tell you the problem with smart security cameras is that they don't know what they're seeing. Those motion pings you get with other products? Defined by how light shifts in front of its sensor, treating an approaching figure or low-flying bird with equal alarm. So, Psync's focus is to improve machine vision, but to also go one step further and pair this vision with GPT-enabled generative AI to help it, and you, understand what it can see. Its first product, the Genie S, is a security camera that'll send you a written description of what (it thinks) is going on.
On paper, the Genie S has a similar feature set to plenty of other affordable home security units I could mention. There's a powered pan-tilt base, five megapixel camera (outputting 2K Video), four LEDs, a microphone and speaker. But there are differences, like the fact it's in the shape of a cube that, when it's not activated, points the lens toward the base. Psync says it's the most compact camera in its class, but probably not by as big a margin as the company hopes. Setup is easy enough - put it on a table, or use the screw mount to place it somewhere more esoteric, plug in the six foot long USB-C cable, and you're on your way.
Psync says that a smarter camera will be better-equipped to capture what's going on at home, but that's not its best use case. VP of marketing Echo Wong says that the hardware is able to record those memorable moments that fly by quicker than we can pull out our phones." But I don't think you would want to buy this on the off-chance it catches junior's first word or steps. The more prosaic sales line, the one that probably wouldn't fly as well, is that it's a security camera with the added promise of not bugging you with needless pings because of the promise of AI smarts.
Buy one of these, and you'll get the choice of a unit with 32GB built-in storage for $35 or 64GB for $40. I mention this up-front because we're very much in you get what you pay for" territory in terms of the picture and sound quality. It shoots vertically-oriented 2K video but the clips are pretty fuzzy, even if you can zoom in to get some halfway useful detail if required. It doesn't like too much light, so if it's pointed at a window (and/or anything reflective) then chunks of the image will get blown out. Similarly, the sound quality is something of a throwback to an earlier age of crunchy, over-compressed streams. You'll get similarly crunchy audio using the talk feature, which has similarly walkie-talkie" vibes that you won't find on pricier hardware.
Of course, that's not what anyone is here for, but to see what this new company - of which little is known -has cooked up with AI. ViewSay is Psync's transcription tool which uses GPT, a form of generative AI, to essentially let the camera describe in text what it's seeing. ViewSay, which currently costs 99 cents a month, promises to identify objects, sort events that triggered the recording in a visual timeline, let you search through the clips with text and, of course, the aforementioned written pings. Pay, your fee, set this up, and your phone will ping when it spots something interesting, and give you the best description of what is going on that it can manage. Users can also set specific categories, like "Person," "Vehicle," "Pet" and will eventually be able to craft tailored alerts, like "a dog jumps on the couch" for alerts.
Oh, but there is a catch - because that fairly reasonable 99 cents a month is just a limited-time trial, before leaping up to $7 a month. Which, we can all agree, is more than a little bit too much to spend on a product like this, especially in this economic climate.
ViewSay is currently in beta, and while the app splash to get you to sign up promises plenty, the company is keen to keep expectations low. My impression so far is that while Psync has the bones of a workable idea here in theory, the nitty-gritty of practice isn't. I pointed the camera at a neutral corner in my office and play-acted in front of it to see what it would do. My fake phone-call, where I learned that my (fictional) wife had discovered the secret to perpetual motion, went unremarked upon and undocumented. Well, kinda - the camera pinged my phone to say that A man is sitting in a chair in a room, looking at his reflection in a mirror."
Actually, I'm being unfair - since the system can also make fairly accurate guesses at other times. Like, while I was setting the hardware up late one evening, I got a ping to tell me that a man is sitting on the floor, holding a cell phone in his hand." A few days later, I pointed the camera at a TV which was turned off, and the Echo Show that was in front of it. I then turned back to use my laptop - which I think was only really visible in the reflection on the TV's screen. Not long after, the app pinged to say it could see a A man is sitting in front of a laptop, looking at the screen, and possibly using it for work or entertainment purposes." Now, this was either a massively-lucky guess, a false positive or a sign of how accurate this will be in future.
When it detects something going on in this manner, the system records a 12-second clip to its local storage. These clips are retained for at least 14 days, and when you've looked at them in the app, you're also able to save them to your phone. I understand you'll also be able to take longer clips when motion is detected but that feature doesn't yet appear to be available. You'll also be able to share a live feed of your camera, using WebRTC, to up to four viewers - through a browser - for up to 30 minutes at a time. You might be wondering about how secure all of this is, and what exactly is happening to your data. Psync told me that its AI model is based on an AWS instance, and the footage is protected using 256-bit AES encryption. The footage recorded will be stored on the device locally, but the initial frame of the video is sent to the cloud for further analysis.
As something of an AI skeptic, and someone who isn't thrilled at wiring up every corner of my home with a camera, I'm by default hostile to Psync's plan here. But I can at least see where Psync is looking to add value to the standard security camera proposition. If you're out and about, and you get a ping saying there's a person in your living room, when there shouldn't be, then that's pretty helpful. Especially if you can just tune into the live feed and see for yourself what's going on and if you need to do something about it. As much as the macro story is scary, I can understand the logic someone would apply to buy one or two of these.
But it's worth saying too that what I just described isn't yet what Psync is selling, only what it is gesturing toward. The system will require more training, and plenty more data from a broader user base, until it can start offering you more concrete descriptions. Now, I'm sure that in a year or two that will be the case, but until then, you're essentially buying into an ecosystem where you're paying for the privilege of being a beta tester.
This article originally appeared on Engadget at https://www.engadget.com/psyncs-genie-s-security-camera-uses-gpt-to-describe-what-it-sees-130043520.html?src=rss