Yasa-1, a multimodal AI assistant, takes on ChatGPT. Waymo expands its robotaxi service in San Francisco. GitHub Copilot's pricing model impact. Google announces new generative AI search capabilities for doctors. The implications of watermarks in AI-generated content.
Read more
AI Summary
Highlights
AI Chapters
Episode notes
auto_awesome
Podcast summary created with Snipd AI
Quick takeaways
DeepMind and academic labs developed a large-scale data set to advance robotic learning.
Protesters voice concerns over open sourcing large language models.
Microsoft and OpenAI are exploring the development of their own AI chips.
Google Cloud introduces AI-powered search capabilities for healthcare.
DeepMind's Reptile AI model is now available as an open-source language model.
TSMC experiences AI chip demand offsetting sales slump.
Deep dives
DeepMind and many others create a large-scale robotic data set
DeepMind, together with 33 academic labs, has created the open X-embodiment data set, consisting of demonstrations from 22 different robot types. The data set includes 500 skills, 150,000 tasks, and more than 1 million episodes. The data set has led to significant performance improvements in robotics models, such as DeepMind's RT1x and RT2x, resulting in a 3x increase in skill capability and closer to 80% success rate in emergent skill evaluation.
Protesters raise concerns about open source AI models
Protesters outside Meta's office voiced their concerns about the open sourcing of large language models, citing their potential danger and lack of control over their usage. The protesters advocate for companies to adopt best practices for providing safe access to models and encourage structured access instead of open sourcing doomsday-threatening models.
Microsoft and OpenAI consider developing their own AI chips
Both Microsoft and OpenAI are reportedly exploring the development of their own AI chips to reduce their reliance on companies like NVIDIA. This move aims to increase competition, negotiate better deals, and provide tailored AI capabilities. It reflects a broader trend of companies prioritizing control over their AI infrastructure.
Google Cloud introduces AI-powered search capabilities for doctors
Google Cloud has announced new AI-powered search capabilities for the healthcare industry, allowing doctors to access and search clinical information from various medical records in one place. This aims to save time for healthcare workers and improve efficiency.
DeepMind's Reptile AI model now available on Hugging Face
DeepMind's Reptile AI model, focused on code completion and assistance, has been released on Hugging Face as an open-source language model. It is now free for all users, with basic AI features available for free and more powerful AI models exclusively for pro users.
TSMC sees AI demand offset slump in sales
Taiwan Semiconductor Manufacturing Company (TSMC) has experienced a slump in sales, but the demand for AI chips has helped offset the decline in other chip markets. Data centers driving AI demand have contributed to the company's continued success, as AI chips and associated memory remain in high demand.
Microsoft may unveil its own AI chip
Microsoft is reportedly planning to unveil its first AI chip at its annual developers conference. The move aims to reduce Microsoft's reliance on NVIDIA-designed chips and gain more control over its AI infrastructure. This follows Google, Tesla, and other companies exploring custom chip designs for AI.
OpenAI researchers develop 'StableLM' language model
OpenAI researchers, in collaboration with others, have developed 'StableLM,' a three-billion-parameter language model. This open-source model aims to improve general-purpose robotics by enhancing skill capability. StableLM has shown improvements in performance and generalization across a range of tasks.
Google's DeepMind collaborates on large-scale data set for robotic learning
DeepMind, in collaboration with 33 academic labs, has developed a large-scale data set called 'openX-embodiment.' This data set contains demonstrations from 22 different robot types and includes over 1 million episodes, 500 skills, and 150,000 tasks. The data set aims to advance robotic learning and improve the generalizability of models across various robot types.
Robotics researchers improve robot learning with diverse data set
Researchers from DeepMind and multiple academic institutions have created a large-scale data set with demonstrations from 22 different types of robots. This open-source data set, called openX-embodiment, consists of 1 million episodes, 500 skills, and 150,000 tasks. Using this data set, researchers have achieved significant performance improvements in robotic learning, showing the promise of general-purpose robotics.
AI servers could consume as much electricity as an entire country by 2027
A peer-reviewed analysis estimates that AI servers could consume between 85 to 138 terawatt hours of electricity by 2027, similar to countries like Argentina, the Netherlands, or Sweden. This raises concerns about the strain on global electricity production and infrastructure.
Writers Guild of America strikes end with new deal that includes AI restrictions
The Writers Guild of America (WGA) has reached a new deal with studios, ending months-long strikes. The deal includes restrictions on AI use, preventing AI from being listed as a co-author or credited as an author, and addressing concerns about AI-generated content in the industry.
Researchers demonstrate that AI-generated deepfakes can bypass watermarks
Researchers have shown that AI-generated deepfakes can be modified to remove watermarks placed on images and can even add watermarks to human-generated images. This raises concerns about the reliability and effectiveness of using watermarks to detect deepfakes.
Disney's 'Loki' poster receives backlash over reported use of generative AI
Disney's TV show 'Loki' faced controversy over a poster that was believed to have been generated using AI. However, it was later discovered that the image was purchased from a stock site and not AI-generated. This incident highlights the need for transparency in the use of AI-generated content.
