The Morning After: Google claims ‘unprecedented photorealism’ from its new text-to-image AI
Google has shown off a new artificial intelligence system that can create images based on text input. Its Imagen diffusion model, created by the Brain Team at Google Research, offers “an unprecedented degree of photorealism and a deep level of language understanding.”
This isn’t the first time we’ve seen AI models like this. OpenAI’s DALL·E (and its successor) performed similar witchcraft, turning text into visuals. Google’s version, however, tries to create more realistic images. The researchers created a benchmark and asked humans to assess each image from a range of AIs. They “prefer Imagen over other models in side-by-side comparisons, both in terms of sample quality and image-text alignment,” Google said.
It’s not available to the public, and there are reasons for this. “Datasets of this nature often reflect social stereotypes, oppressive viewpoints, and derogatory, or otherwise harmful, associations to marginalized identity groups,” the researchers wrote. Imagen has inherited the “social biases and limitations of large language models” and may depict “harmful stereotypes and representation.” The team said the AI encodes social biases, including a tendency to create images of people with lighter skin tones and place them in certain stereotypical gender roles. The system could be used to make unsavory images to intentionally cause offense.
The team may eventually allow the rest of us to play with the model to generate our own images, but the researchers need to consider a framework first — a challenge in itself.
— Mat Smith
The biggest stories you might have missed
Apple’s latest Pride Edition Watch bands include a nod to the company’s history
Linksys rolls out a pair of more affordable WiFi 6 mesh routers
Google is testing a smaller, modular Street View camera system
Samsung’s new Smart Monitor M8 is $100 off for the first time
‘Lord of the Rings: Gollum’ hits consoles and PC on September 1st
Microsoft’s Project Volterra is a mini PC for ARM developers
Panasonic’s latest modular Toughbook is larger and lighter
Harley-Davidson made an electric mountain bike without front or rear suspension
Starlink launches its satellite internet for RVs
But it costs more than a standard Starlink service.
Starlink’s internet service for RV drivers and ‘vanlife’ types is live. While applying for a regular Starlink dish and service will put customers on a waitlist until 2023, Starlink for RVs is immediately available and will ship out to buyers right now. However, network resources are always deprioritized for RV connections, and the service costs $135, which is $25 more than a regular Starlink connection.
ASUS’ ROG Swift is the ‘world’s first’ 500Hz G-Sync gaming monitor
It uses NVIDIA’s Reflex tech to reduce system latency.
ASUS has unveiled what it calls the “world’s first” 500Hz G-Sync gaming display, the 1080p ROG Swift 500Hz. Designed for competitive gaming, it uses a special panel and incorporates NVIDIA’s G-Sync Esports technology to maximize motion clarity. It also uses NVIDIA’s Reflex Analyzer technology, which delivers real-time stats to help you reduce end-to-end latency if you’re using a Reflex-optimized mouse and NVIDIA GPU.
The key highlight remains the 500Hz refresh rate, which draws eight times faster than typical 60Hz displays.
Canon’s EOS R7 and EOS R10 are its first EOS R crop-sensor cameras
Well-priced and fast but lacking backside-illuminated/stacked sensors.
Canon has launched its first EOS R APS-C crop sensor cameras, the 32-megapixel EOS R7 and 24-megapixel EOS R10. The new models bring Canon’s APS-C and full-frame RF series in alignment, so you can finally use lenses interchangeably. More importantly, they carry impressive specs, like 15 fps mechanical shutter shooting speeds and 4K video at up to 60 fps.
The EOS R10 will cost $980 for the body only and $1,100 with the RF-S18-45mm lens and $1,380 with the RF-S18-150mm lens. The R7, meanwhile, will sell for $1,500 for the body only and $1,900 with the S18-150mm lens. Both should arrive later this year.
Microsoft Teams goes beyond screen sharing for real-time collaboration
Live Share is a major evolution for remote work.
Microsoft’s new Live Share feature should make it easier for Teams apps to enable real-time collaboration. If this sounds familiar, it’s because Microsoft announced plans to make Teams the go-to option for collaborative apps last year. Live Share is based on Fluid Framework, Microsoft’s attempt at atomizing components of traditional documents and making them collaborative. Microsoft says several partners, including Accenture, Frame.io and Hexagon, are already building Live Share experiences in Teams projects.
NVIDIA is bringing out liquid-cooled GPUs
But they’re for data centers.
Later this year, NVIDIA will begin selling a liquid-cooled version of its A100 GPU for data centers. The GPU maker is positioning the video card as a way for cloud computing companies to make their facilities more energy-efficient. NVIDIA claims a facility outfitted with its water-cooled A100 GPUs ran the same workload as an air-cooled data center while using about 30 percent less power.
Google has shown off a new artificial intelligence system that can create images based on text input. Its Imagen diffusion model, created by the Brain Team at Google Research, offers “an unprecedented degree of photorealism and a deep level of language understanding.” Google This isn’t the first time we’ve seen…
