Tesla AI Day 2022 | Google Showcases PaLI

Key takeaways

• Tesla AI Day 2022 unveiled the first iteration of Tesla's robot.
• OpenAI announces new flexibility for Dall-E 2 AI, allowing users to upload faces with consent and generate variations on the pictures.
• Scientists develop an AI laughing robot named Erica to recreate different types of laughter.
• AmazeVR raises $17m in funding to create immersive VR concerts.
• Google showcases PaLI, a multilingual AI system that performs tasks in over 100 languages.

Tesla AI Day 2022 unveiled the first iteration of Tesla's robot

During Tesla AI Day 2022, Elon Musk introduced the first iteration of Tesla's robot, which showcased its walking capabilities before leaving the stage. The event featured presentations from various members of Tesla's AI and hardware teams, displaying videos of the robot carrying out tasks such as lifting objects and watering plants. The aim of the demo was to exhibit progress and demonstrate Tesla's direction, inspiring confidence and attracting talent to further develop the program. Musk revealed that the initial prototype, named Bumble C, will evolve into Optimus, an advanced robot capable of efficient walking, balance, carrying heavy loads, using tools, and having precise grips. The Tesla bot incorporates elements from Tesla's energy products, including battery management and the supercomputer used in Tesla vehicles, as well as features from the Autopilot driver assistance system, such as wireless connectivity and hardware-level security.

OpenAI announces new flexibility for Dall-E 2 AI, allowing users to upload faces with consent and generate variations on the pictures

OpenAI's Dall-E 2 AI allows users to upload faces for the first time, relaxing its rules while maintaining safeguards against inappropriate content, amid growing competition in the image-generating sector. Initially, generating images with realistic faces was banned during the public beta launch, but later, it was allowed without depicting specific individuals. Now, users can upload photos of real people with consent and generate various image variations using the AI tool. OpenAI communicated the rationale behind these changes to users in a letter, stating that they have improved the technology to prevent the generation of inappropriate content.

Scientists develop an AI laughing robot named Erica to recreate different types of laughter

Scientists are working on an AI system named Erica, a laughing robot designed to emulate the nuances of humour and respond with the right type of laughter at the right moment, improving natural interactions between humans and AI. The machine learning system was trained using annotated dialogue data for different types of laughs, such as solo laughs, social laughs, and laughs of mirth. By mimicking appropriate types of laughter based on audio files, the algorithm learned the characteristics of social and mirthful laughs, aiming to reflect them in suitable situations. This ability to laugh appropriately may lead to the development of robots with unique and empathetic characters, enhancing their interactions with humans.

AmazeVR raises $17m in funding to create immersive VR concerts

Los Angeles-based virtual concert platform, AmazeVR, secures $17m in funding to create immersive music experiences through virtual reality (VR) concerts, aiming to capitalise on the growing popularity of virtual shows in the entertainment industry, which gained momentum during the pandemic. Co-CEO Steve Lee plans to use the financing to expand partnerships with artists, management agencies, labels, and publishers, targeting top artists in the U.S. AmazeVR is preparing virtual concerts and a music metaverse service for major VR app stores, compatible with next-gen headsets like Meta Quest Pro and Apple's rumoured VR headset, with the goal of broadening content diversity and attracting VR concert fans worldwide.

Google showcases PaLI, a multilingual AI system that performs tasks in over 100 languages

Google researchers have showcased PaLI, an AI system capable of performing tasks in over 100 languages, addressing the historical limitations of single-language AI systems like OpenAI's GPT-3. The recent acceleration in multilingual system research led to PaLI's creation, trained on images and text for image captioning, object detection, and optical character recognition. Google asserts PaLI's comprehension of 109 languages and their word-image relationships, enabling it to perform in various languages. While still in the research phase, the team sees the potential significance of the interaction between language and images, hinting at the possibility of a future commercial product.