SaraEye - This is the world's first ChatGPT with a sense of sight! 

Exactly right. The cameras observe the environment, and in the background, a separate program identifies objects, people, motion, etc. This information is invisibly sent as prompts to ChatGPT, which then responds as you can see in the video.

​@@ArturMajtczak This reminds me of the SHRDLU program developed by Terry Winograd at MIT in 1968 (based on around fifty nouns, verbs and adjectives in 3D world of blocks). I guess that it takes a lot of computing power to do this analysis, and that you do it on an external server from the pairs of images sent. I guess that Rasperry and your SaraKit card are used only to position the motors, process the image and sound and communicate with the server, is it ? Your approach is interesting. Do you think the GPT-4 Vision update uses a similar principle and communicates through prompts with the conversational system? (this system quickly has its limits it seems to me).

@@monsieur3d985 Sending images to a server and waiting for a response is indeed too slow and costly, so the image analysis is actually done on the Raspberry Pi itself, using a simple trained model. While this model might not recognize everything, it certainly has broad and sufficient capabilities. Image recognition isn't performed in real-time at 25 frames per second - that's not necessary at this stage. We just analyze changes in the background image, which takes about 100 to 600 ms. As I mentioned, this process runs in a separate thread and is efficient enough for our purposes. In terms of the GPT-4 Vision update, while it might use a similar principle of communicating with the conversational system through prompts, our approach focuses on local processing to avoid the delays and costs associated with server-based processing. This method, although it has its limitations, is quite effective for our current needs.

We are developing the project, but it is not available for sale yet

Kind of ironic that you call this a "natural" conversation. It is so UNnatural that I'm not sure which is the robot -- the little black thing sitting on the desk or the bigger white thing sitting on the chair.

@@2000ReRyRo heh heh

1. You don't need to use a wake word like 'Alexa' or 'OK Google'; just look at SaraEye, it sees you and knows you are speaking to it - it’s more natural, the way people communicate. When we are in a group and speak while looking at someone, that person knows we are speaking to them... 2. By looking at the device, and more importantly, SaraEye looking back at you, a unique bond is formed that is hard to achieve by talking to a 'speaker' like Alexa. :)