AI News – Week of 9/3/2023
Google Meet’s new AI will be able to go to meetings for you
·????? Google Meet is introducing new AI-powered features, including a tool called Duet AI that can attend meetings on your behalf and take notes in real time.
·????? Users can click "Attend for me" on a meeting invite, and Duet AI will generate talking points that are viewable to attendees during the meeting. This feature aims to be a time-saving tool, especially when users are double-booked or unable to attend a meeting suddenly.
·????? Another feature allows users to catch up on missed parts of a meeting by providing a mid-meeting summary and facilitating private chats with a Google chatbot to go over missed details.
·????? The AI tools are designed to free people from the task of taking meeting notes and to facilitate easier catch-up during and after calls. However, the effectiveness of these tools is yet to be determined, as AI can sometimes make mistakes.
·????? These features are expected to roll out in the coming months, with some arriving next year. Google aims to enhance the video conferencing experience by integrating AI into its products and adding dynamic layouts to Meet.
·????? Google's senior director of product for Meet, Dave Citron, highlighted the evolution of video conferencing and mentioned that the company is at an inflection point where they are incorporating large language models and diffusion models to innovate their products.?
?
Google releases Beta of SGE - their generative search
Supercharge your search to find what you’re looking for in faster, easier ways. Get AI-powered overviews to quickly get the gist of a topic and easily follow up. Let AI do the heavy lifting for you.?
?
Google Workspace adds New AI features
·????? Google has announced a series of generative AI features to be integrated into its workspace apps including Google Docs, Gmail, Sheets, and Slides.
·????? These features will facilitate generating, summarizing, and brainstorming text in Google Docs, akin to OpenAI's ChatGPT, and allow the creation of AI-generated imagery, audio, and video for presentations in Slides.
·????? The initiative is a part of Google's strategy to compete with Microsoft and other rivals in the AI domain. Google is accelerating the integration of AI tools into its products, with some features being tested by a group of trusted users.
·????? The new AI functionalities will assist in various tasks such as drafting emails in Gmail based on brief bullet points, creating job descriptions in Docs, and generating insights and analysis in Sheets.?
?
Apple has significantly increased its investment in conversational AI
·????? Despite skepticism from Apple's AI chief, John Giannandrea, a team has been working on conversational AI for the past four years.
·????? The company is experimenting with large language models, with some employees having access to an internal chatbot named AJAX, which is more advanced than the original ChatGPT-3.5.
·????? Apple aims to develop features allowing Siri to automate multi-step tasks, with potential integration in iOS 18.
·????? Apple is also working on software to generate videos and images, and multimodal AI that interacts with images, video, and text.?
?
Claude LLM Model increases to 100,000 token context window
·????? This enhancement allows Claude to analyze and digest hundreds of pages of material in less than a minute, facilitating prolonged conversations that can last for hours or even days.
·????? The 100k context window can be utilized to digest, summarize, and explain dense documents such as financial statements, research papers, and legal documents, as well as analyze strategic risks and opportunities based on company annual reports.
·????? It can also assist in identifying the pros and cons of legislation, extracting information from developer documentation, and rapidly prototyping by analyzing entire codebases.
·????? A demonstration of this feature was conducted by AssemblyAI, where a long podcast was transcribed into 58k words and then used Claude for summarization and Q&A.
·????? The 100k context windows are now available in their API, and more details can be found on their website for those interested in utilizing Claude.
·????? The announcement emphasizes the efficiency of Claude in processing large amounts of text, highlighting a case where it identified a modified line in "The Great Gatsby" in 22 seconds.?
?
TII Falcon 180b Model Released
·????? As of September 2023, Falcon 180B is the highest-performing pre-trained large language model on the Hugging Face open leaderboard.
领英推荐
·????? The model requires substantial computational resources for inference, needing 640GB of memory and eight A100 80GB GPUs when quantized to half-precision (FP16), or eight A100 40GB GPUs (320GB memory) when quantized to INT4. This setup can cost around $20,000 per month to maintain.
·????? Falcon 180B allows commercial usage and offers organizations more control and ownership over their model compared to alternatives like OpenAI's GPT-4.
·????? Performance-wise, Falcon 180B is impressive, outperforming GPT-3.5 on some benchmarks and scoring between GPT-3.5 and GPT-4 on most benchmarks.
·????? The page indicates that more information and guides about Falcon 180B will be added in the future.?
Robots Play Tag
·????? Researchers at UC Berkeley have developed a new method to teach robots strategic decision-making for dynamic tasks like playing tag, where robots can now chase humans more effectively.
·????? The new approach, termed "privileged learning," involves a robot teacher guiding a student robot step-by-step based on the future trajectory of the person or robot being chased, making the complex planning problem a simpler supervised learning task for the student robot.
·????? The robot learns dynamic behaviors such as reducing its speed when the person or robot being chased turns, or predicting where they will be and intercepting them.
·????? The method has been tested successfully on a real four-legged robot that played tag with humans and other robots, relying solely on built-in cameras and proprioception.
·????? Despite the success, the system currently cannot handle obstacles and requires more extensive AI training and better sensors to improve further.?
?
AI Hearing Aids
·????? The article discusses the integration of AI technology in hearing aids, offering features like real-time language translation and fall detection.
·????? A Minneapolis-based company, Starkey, is at the forefront of this innovation, aiming to change the perception of hearing aids not just as a tool for the elderly but as a device offering "superpowers" like language translation and environmental adaptation.
·????? The AI-powered hearing aids can act as a personal assistant, allowing users to take phone calls, listen to music, and even alert up to three people if the user has a fall.
·????? These devices can recalibrate up to 80 million times an hour, adapting to different listening environments automatically.
·????? The company aims to attract younger users to the product, with features that even appealed to the CEO's 12-year-old son, who used it for language translation in a French class.
·????? The article highlights the potential of AI in enhancing the quality of life for individuals with hearing loss, and possibly predicting falls before they happen or integrating other health monitors in the future.?
?
How a Brain Implant and AI Gave a Woman with Paralysis Her Voice Back?
This video showcases groundbreaking technology developed by researchers from UC San Francisco and UC Berkeley that helps individuals with paralysis communicate more naturally through a digital avatar.
·????? The technology decodes brain signals of a woman named Ann, who has locked-in syndrome due to a stroke 18 years ago, converting them into text, speech, and facial expressions for a digital avatar.
·????? The technology involves an electrocorticography grid attached to the participant's skull, which detects neural activity related to attempts to move facial muscles, translating them into speech.
·????? Ann demonstrates a conversation with another individual using this technology, showcasing a significant improvement in communication speed compared to her old assistive communication device.
·????? Ann aspires to use this technology to become a counselor and help others, emphasizing the importance of developing technologies to support individuals with disabilities and integrate them into workplaces.
·????? The research team, led by UCSF's Edward Chang, MD, has been working towards this goal for the last 10 years, aiming to restore speech and provide a real solution for patients.
Da Vinci robot surgeon removes inoperable tumor, saving patients life?
A retired CBC journalist, Glenn Deir, who was facing his second major bout with cancer, had a life-saving surgery performed with the help of the Da Vinci surgical robot. The tumor, located on his tonsils and extending to the back of his tongue, was initially deemed inoperable. Traditional surgery would have required cutting his jaw in two to remove the tumor manually, a procedure that no surgeons were willing to undertake.
The Da Vinci robot, controlled by Dr. Martin Corsten, was able to reach the tumor with an incision on the neck, avoiding the need to cut the jaw. Despite complications, including a stiff tonsil due to previous radiation and a large tumor on the tongue, the surgery was successful.?
Deir is now in the recovery phase, learning to speak again after having parts of his tongue, throat, and tonsil cut. This case highlights the significant advancements in surgical robotics, showcasing their potential to perform complex surgeries that might have been deemed too risky or impossible in the past.?
?