Google had a plethora of AI announcements at Google I/O.

Gemini Advanced subscribers now have access to the newest model, Gemini 1.5, which has a 1-million-token context window. In practice the "context window" is the combined size of your prompt and any underlying "system message" that the creators of the system put in.

Google demoed a "Ask your photos" feature where you can ask questions like "What's my license plate number" and it searches all your photos and finds your license plate number and tells you what it is. You can ask when your kid learned how to swim. You can ask questions of your Gmail, such as "summarize all the announcements from my kid's school".

Google is working towards AI agents that will do multiple steps for you instead of just answering one question. You could tell it to complete a task for you and then it will go and try to complete all of those steps to complete the task. "Return these shoes for me." It figures out where the shoes came from, how much they cost, how to contact customer support, and then it actually contacts the shoe seller.

Their lightweight model is called Gemini 1.5 Flash and is designed to run on mobile phones.

Project Astra is their attempt to create a real-time AI agent that uses the camera on your phone. You can ask it to explain what you're looking at, "What is this part of the speaker called?" or ask it to make up rhymes.

Google's response to OpenAI's Sora is a video generation model called Veo.

Google is rolling out an "AI Overview" in Google Search. (I've already seen it.) It uses what they call "multi-step reasoning". You should be able to ask Google Search "multi-step questions".

They're building AI into Android phones that can detect if you're potentially talking to a scammer.

They're open-sourcing a 2-billion parameter model called Gemma 2.

Google just took over the AI world (a full breakdown) - Matt Wolfe

#solidstatelife #ai #genai #llms #multimodal #edgecomputing #google

Aethero is a new startup promising "edge computing" AI systems in satellites. "The next generation of space rated edge computing".

"Aethero's onboard software includes a thin, headless system containing a minimal set of packages and tools needed to boot the system. The onboard system will allow you to run containerized applications and will provide services such as Over the Air software updating. The Over the Air testing or updating and fleet management or system level testing is accessed through Aethero's unified Aether Software. We provide an automated framework for platform testing; it includes support for the Hardware, Board Support Packages and Software -- this allows users to develop, debug and test multi-node device systems reliably, scalably and effectively."

"Users can use Aethero's AMATDT (Automated Model Annotation, Training & Deployment Tool) that is integrated within the Aether Software to customize models deployed on Aethero's Edge Computing Modules. Current imagery support includes RGB, Multispectral, and Hyperspectral data."

This is tailored to Aethero's hardware.

"Aethero is leveraging standard architectures such as the CubeSat or PC104 framework, state-of-the-art radiation-hardened commercial components, and modern software operating systems to provide a high performance, highly capable single board computers, or multiple redundant or distributed computing configurations. The versatile architecture means that the system can be used for a variety of applications such as Autonomous Spacecraft Operation, Machine Vision Operations such as [VPS] Visual Positioning Systems or Optical Navigation, Imagery Processing ([UV/EO/IR] Ultraviolet/Electro-Optical/Infrared, [MSI] Multi-Spectral, [HSI] Hyperspectral, [SAR] Synthetic Aperture Radar, [LIDAR] Light Detection and Ranging, [TIR] Thermal Infrared), [RF] Radio Frequency Signal Processing such as Link-Budget Optimization, Video/Image processing such as Manipulation or Segmentation, Object Detection with Classification, [AI] Artifical Intelligence/[ML] Machine Learning Applications, [SDR] Software Defined Radio, Data Compression/Management, etc."

Aethero -- space data, re-imagined

#solidstatelife #ai #edgecomputing #aihardware #space #satellites

The Velo AI smart bike light.

"In moving from radar to a camera-based solution, the aim was to create a device that could tell the cyclist 'a lot more about what's going on in the world and do a lot of things that the radar can't do'. The device would need to help cyclists by providing situational awareness and alerts about nearby vehicles. This includes the ability to distinguish, using computer vision algorithms, between different vehicle types, as well as to estimate their relative speed and to identify and predict driver behaviour."

"Raspberry Pi Compute Module 4 is in effect the brain of the Copilot, aided by a custom Hailo AI co-processor to run the neural networks required for the device's computer vision. A fixed-lens Arducam camera is used to record video footage."

"The Copilot is supplied with a mount to fix it to a bike's seat post or saddle rail, with the camera facing rearward. The AI analyses the live video footage and, depending on the type of driver behaviour detected, custom alerts may be triggered -- audible for the cyclist, and flashing LED light patterns to alert the driver behind."

The article doesn't mention it, but "velo" means bike in French.

Velo AI smart bike light - Raspberry Pi

#solidstatelife #ai #computervision #edgecomputing #raspberrypi

"Offline Chat: Private AI". A $2 iPad and iPhone app, Mistral 7B 0.1 quantized to 3-bit. That sounds like an impressive technical feat but it didn't get good reviews.

Offline Chat: Private AI

#solidstatelife #ai #llms #edgecomputing

NVIDIA Unveils Jetson AGX Orin Developer Kit

#artificialintelligence #news #edgecomputing #jetson #nvidia #orin #hackaday
posted by pod_feeder_v2