Quality VR gaming, AI superpowers for Quest 3 & Meta's GPT-3 moment
Our weekly recap: Captivating VR games do exist. And Meta's new image segmentation model could help Quest 3.
New and unknown, yet charming
In addition, Half-Life 2 and Black Mesa are now fully playable in VR, and just last week we had the addictive Ghost Signal Stellaris.
Avatar as AI Chatbot
Chatbots like ChatGPT are taking over screens. They could also make video game characters more believable. The AI caddie "Arthur" demonstrates the enormous impact that the fluid and flexible speech capabilities of VR avatars could have. For example, Arthur can remember the player's name and use it in conversations. He can also automatically create putting challenges and drills that require only information such as putting distance and terrain.
Next-Generation Computer Vision for Quest 3
Meta's "Segment Anything" (SAM) is a new image analysis model that can accurately detect and distinguish objects and content in images. When applied to Quest 3's pass-through AR mode, Meta's SAM could enable higher quality and more complex AR applications by helping the VR headset better understand its environment. For example, SAM could recognize baking accessories on an AR headset and automatically display appropriate tips.
Pulse 3D Headset for PSVR 2 Review
Are the Playstation VR 2's in-ear headphones enough? Sony's wireless headset proved our editor Tomislav wrong. His conclusion after testing the Pulse 3D Headset for PSVR 2: The sound is worlds better and reminds you how much audio contributes to immersion.
The GPT-3 moment for Computer Vision
As mentioned above, Meta may have hit the big time for AI-based image analysis with Segment Anything. It could serve as a central building block for future AI applications, and thus be something of a GPT-3 moment for computer vision, since the model can be adapted to many tasks - including XR - with just a few examples.
In other AI news at THE DECODER, MemoryGPT is a demo for a ChatGPT with long-term memory that can remember and retrieve dialogues many years old, and models like Auto-GPT could make large language models autonomous.