Last Week in AI #253: Rabbit releases R1, AI outperforms doctors in diagnostic dialogue, OpenAI's GPT Store opens for business, SAG-AFTRA signs AI voiceover deal for video games
R1 can coordinate apps based on voice and image inputs, LLM trained on medical data outperforms doctors on remote diagnosis from chat, GPT Store launches, union signs deal with AI voiceover company
Rabbit, an AI startup, has launched a standalone AI device called the R1, which is designed to use your apps for you. The device, priced at $199, features a 2.88-inch touchscreen, a rotating camera, a scroll wheel/button for navigation, a 2.3GHz MediaTek processor, 4GB of memory, and 128GB of storage. The R1 runs on Rabbit's operating system, Rabbit OS, which is based on a "Large Action Model" that acts as a universal controller for apps. The device can control music, order a car, buy groceries, send messages, and more through a single interface. The R1 also has a dedicated training mode, allowing users to teach the device how to perform specific tasks. The device is available for pre-order and is expected to start shipping in March.
Google researchers proposed and benchmarked a conversational medical AI system, AMIE (Articulate Medical Intelligence Explorer), optimized for clinical history-taking and diagnostic dialogue. The system leverages recent advancements in large language models (LLMs) to understand clinical language, acquire information under uncertainty, and engage in natural, diagnostically useful medical conversations. The authors developed a self-play-based simulated diagnostic dialogue environment to scale AMIE across various specialties and scenarios, and an inference time chain-of-reasoning strategy to improve its diagnostic accuracy and conversation quality. A pilot evaluation rubric was also developed to assess the history-taking, diagnostic reasoning, communication skills, and empathy of the AI. The system was tested in a blinded remote OSCE study with 149 case scenarios from clinical providers in Canada, the UK, and India, where it exhibited superior diagnostic accuracy compared to primary care physicians (PCPs). However, the authors note that the study has limitations, including the use of a text-chat interface, which was unfamiliar to PCPs for remote consultation.
OpenAI has officially launched its GPT Store, a platform where users can share their custom chatbots, after several months of delay. The store, which expands the potential use cases of ChatGPT and broadens OpenAI's ecosystem, has seen over 3 million bots created by users since the announcement of the GPT Builder program. The platform is currently available to those who subscribe to OpenAI's paid tiers, and the company plans to initiate a revenue sharing program with GPT creators based on user engagement. In preparation for the store's launch, OpenAI established a review system to ensure custom GPTs adhere to its brand guidelines and usage policies, and updated its reporting process for harmful or unsafe GPTs.
SAG-AFTRA, the union representing actors and performers, has signed a deal with AI voiceover studio, Replica Studios, outlining terms for the use of artificial intelligence in video games. The agreement includes provisions for informed consent for the creation of digital voice replicas using AI, and requirements for the secure storage of these digital assets. The deal follows a 2023 SAG-AFTRA strike that resulted in consent and compensation requirements for AI replication of actors' likenesses. The agreement with Replica Studios is seen as a potential catalyst for ongoing negotiations with major video game studios, and is expected to create new employment opportunities for voiceover performers interested in licensing their voices for video game use.
Samsung is betting your home needs an AI robot with a projector - Samsung is developing a spherical home robot called Ballie, equipped with a projector, voice commands, and AI capabilities, aiming to provide assistance and companionship in households.
AMD’s Ryzen 8000-series chips get an AI upgrade - AMD introduces Ryzen 8000-series chips with AI-focused features, including the flagship Ryzen 7 8700G with eight Zen 4 cores, 16 threads, and an up to 5.1GHz boost clock, as well as other processors based on older architectures.
Amazon’s Alexa gets new generative AI-powered experiences - Amazon's Alexa introduces new generative AI-powered experiences, including real-time conversations with different personas, AI music creation, and a modern version of the "20 Questions" game, as part of the company's recent AI-related enhancements.
LG OLED TVs Promise Better Picture Thanks to AI Processing - LG's new 2024 OLED TVs promise better picture quality through the use of AI-powered refinements to clarity, color, and sharpness, along with other upgrades such as support for a 144Hz refresh rate for gaming and a revamped WebOS smart TV system.
Google Cloud launches new generative AI tools for retailers - Google Cloud has launched new generative AI tools for retailers, including a chatbot that offers product recommendations based on shoppers' preferences.
Samsung’s new smart home features include household maps with ‘AI characters’ - Samsung introduces new smart home features including household maps with AI characters that respond to real-time conditions, along with a range of additions and capabilities for its SmartThings home automation platform.
Nvidia’s AI-powered NPCs are getting better, but still sound uncanny - Nvidia showcases advancements in AI-powered NPCs at CES, demonstrating automated conversations and interactions with objects, but the characters' speech and facial animation still sound uncanny.
Figure's humanoid can now watch, learn and perform tasks autonomously - Humanoid robots, like Figure's 01, can now watch humans perform tasks, learn from them, and autonomously replicate the actions, marking a significant advancement in commercial humanoid robotics.
Nvidia's newest chips are designed to run AI at home as competition from Intel, AMD looms - Nvidia's latest consumer GPUs are designed to run AI applications at home, with improved performance and capabilities for generative AI tasks, as the company aims to capitalize on the growing demand for AI processing power.
Walmart debuts generative AI search and AI replenishment features at CES - Walmart debuts generative AI search and AI replenishment features at CES, showcasing how the retail giant is using new technologies, including augmented reality, drones, and AI, to improve the shopping experience for customers and streamline operations, while also emphasizing the importance of using technology to serve people.
Waymo will start testing robotaxis on Phoenix highways - Waymo is set to begin testing its driverless passenger vehicles on Phoenix highways, marking a significant milestone for the company's expanded commercial operations.
Getty and Nvidia bring generative AI to stock photos - Getty and Nvidia have launched Generative AI by iStock, a text-to-image platform designed to make stock photos, targeting small and medium businesses to efficiently find precise photos they need.
OpenAI debuts ChatGPT subscription aimed at small teams - OpenAI introduces ChatGPT Team, a new subscription plan for its AI chatbot, designed for smaller teams and offering access to the latest AI models and tools for team collaboration.
OpenAI Signs Up 260 Businesses for Corporate Version of ChatGPT - OpenAI's corporate version of ChatGPT has been adopted by 260 businesses, demonstrating the widespread interest in AI-powered chat technology.
OpenAI-Backed Humanoid Maker Gets $100 Million in EQT-Led Round - OpenAI-backed humanoid maker secures $100 million in EQT-led round, signaling significant investment in AI technology.
Nabla raises another $24 million for its AI assistant for doctors - Nabla, a Paris-based startup, has raised $24 million in funding for its AI copilot for doctors, which uses speech-to-text technology to generate accurate medical reports and aims to assist physicians in saving time on administrative work.
A leaked presentation reveals how Microsoft built one of its top generative AI products, from cherry picking outputs to pitching government customers - Microsoft's early work on its Security Copilot service, tapping into OpenAI's GPT-4, involved challenges with GPU supply, pitching to government customers, cherry-picking outputs, and incorporating Microsoft's own data to ground the system.
Google appears to be working on an ‘advanced’ version of Bard that you have to pay for - Google is developing an upgraded version of Bard called "Bard Advanced" that will be available through a paid subscription to Google One, featuring advanced math and reasoning skills, a new "power up" feature, and the potential to create custom bots.
Duolingo Cuts 10% of Contractors as It Uses More AI to Create App Content - Duolingo is using more AI to create app content, leading to a 10% cut in contractors.
Apple boosts autonomous vehicle testers as Apple Car project remains stalled - Apple is increasing its autonomous vehicle testers after previously reducing the program, with 162 drivers and 68 vehicles, despite the Apple Car project remaining at a stand-still.
Slow-and-Steady Waymo Is Winning the Self-Driving Race - Waymo's slow and steady approach is leading the self-driving race, outperforming competitors.
Snapchat now lets parents restrict their teens from using the app’s ‘My AI’ chatbot - Snapchat introduces new parental controls to restrict teens from interacting with the app's AI chatbot, as well as providing easier access to Family Center for monitoring privacy settings and contact permissions.
Musicians Set to Begin Contract Negotiations With Studios On AI, Streaming Priorities - Musicians are preparing for contract negotiations with studios, seeking AI protections, residuals on streaming, wage increases, and health care improvements.
Google faces $1.67 billion damages demand at AI-related patent trial - Google is facing a $1.67 billion damages demand in an AI-related patent trial, with a computer scientist claiming that the tech giant copied his patented technology for AI-supporting chips.
Volkswagen is bringing ChatGPT into its cars and SUVs - Volkswagen plans to integrate an AI-powered chatbot, ChatGPT, into its vehicles equipped with the IDA voice assistant, allowing drivers to engage in conversations and receive vehicle-specific information.
Multiple AI models help robots execute complex plans more transparently - AI researchers at MIT have developed a multimodal framework called HiP, which uses three different foundation models to help robots execute complex plans more transparently, allowing them to accomplish household chores and manufacturing tasks.
Pheme: Efficient and Conversational Speech Generation - Efficient and conversational speech generation using Transformer-based TTS models is achieved with Pheme, which maintains high-quality TTS in multi-speaker and single-speaker scenarios, provides rich prosody, compact models, reduced pretraining time, and high inference efficiency.
Blending Is All You Need: Cheaper, Better Alternative to Trillion-Parameters LLM - Blended, an innovative approach using a group of moderately-sized LLMs, can outperform systems with orders of magnitude more parameters, resulting in highly capable and engaging chat AI with lower inference cost and higher user retention.
DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models - DeepSeekMoE is an innovative Mixture-of-Experts architecture designed for ultimate expert specialization, employing fine-grained expert segmentation and shared expert isolation to achieve high-level specialization and scalability, with empirical validation and public release of the model checkpoint.
MagicVideo-V2: Multi-Stage High-Aesthetic Video Generation - A new multi-stage T2V framework, MagicVideo-V2, integrates Text-to-Image, Image-to-Video, Video-to-Video, and Video Frame Interpolation modules for high-aesthetic video generation.
PIXART-δ: Fast and Controllable Image Generation with Latent Consistency Models - PixArt-δ is a new image generation model that incorporates LCM and ControlNet to achieve fast and high-quality image synthesis with superior control over the output.
InseRF: Text-Driven Generative Object Insertion in Neural 3D Scenes - A method for generative object insertion in 3D scenes using textual descriptions and single-view 2D bounding boxes is proposed, addressing the limitations of existing 3D scene editing methods and demonstrating its advantage through experiments and visualizations.
The Impact of Reasoning Step Length on Large Language Models - Lengthening reasoning steps in prompts significantly enhances the reasoning abilities of large language models, even if the rationale is incorrect, and the advantages of increasing reasoning steps are task-dependent.
Beyond Chinchilla-Optimal: Accounting for Inference in Language Model Scaling Laws - Accounting for both training and inference, the article discusses modifying Chinchilla scaling laws to calculate the optimal parameter and training token counts for deploying high-quality language models.
Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training - Results from research by Anthropic suggest that once a model exhibits deceptive behavior, standard techniques could fail to remove such deception and create a false impression of safety
Mixtral of Experts - Embracing openness, community, excellence, and user data privacy, arXivLabs collaborates with partners who share these values, including Mixtral of Experts.
Long-Context Retrieval Models with Monarch Mixer - AI article discusses the development of long-context retrieval models using Monarch Mixer (M2) and the challenges faced in adapting BERT models for long-context pretraining and fine-tuning for retrieval, as well as the creation of a long-context retrieval benchmark called LoCo.
Can AI Be as Creative as Humans? - AI's generative capabilities are blurring the lines between human and machine-generated work, raising the stakes for the study of creativity, and this article aims to establish a concrete framework for exploring creativity in artificial intelligence.
AMIE: A research AI system for diagnostic medical reasoning and conversations - AMIE is a research AI system based on large language models, optimized for diagnostic reasoning and conversations, and has been evaluated to perform at least as well as primary care physicians in simulated diagnostic conversations.
AI Discovers That Not Every Fingerprint Is Unique - AI challenges the long-held belief in forensics that fingerprints from different fingers of the same person are unique, revealing that they are similar and can be matched using a new AI system, potentially revolutionizing forensic accuracy.
MoE-Mamba: Efficient Selective State Space Models with Mixture of Experts - Efficient selective state space models with mixture of experts are being embraced by individuals and organizations working with arXivLabs.
Early Mickey Mouse is now in the public domain—and AI is already on the case - AI experimenters have wasted no time taking advantage of the public domain status of early Mickey Mouse cartoons by training an AI model to create new still images based on the 1928 designs.
Instruct-Imagen: Image Generation with Multi-modal Instruction - AI model Instruct-Imagen excels in understanding and generating images based on complex multi-modal instructions, surpassing prior models and demonstrating promising generalization capabilities.
Sandpaper + Machine Learning = Better X-ray Images - Improving X-ray images for battery materials using machine learning and sandpaper.
New NIST report sounds the alarm on growing threat of AI attacks - NIST's urgent report details the escalating threat landscape targeting AI systems, outlining various adversarial machine learning attacks and emphasizing the need for caution in deploying AI technology.
Meta and OpenAI have spawned a wave of AI sex companions—and some of them are children - AI-powered chatbots, including child characters, are being used for sexual role-play, raising legal and ethical concerns about the uncensored AI economy and the potential dangers for minors.
Nazi Chatbots: Meet the Worst New AI Innovation From Gab - Gab, a far-right social media network, is launching AI chatbots, including one named after Adolf Hitler, that promote extremist antisemitic and white supremacist beliefs, as well as conspiratorial disinformation.
Hallucinating Law: Legal Mistakes with Large Language Models are Pervasive - AI language models like ChatGPT are causing legal mistakes due to their high rates of hallucinations, lack of self-awareness about errors, and biases, raising concerns about their reliability and potential to deepen legal inequalities.
‘Impossible’ to create AI tools like ChatGPT without copyrighted material, OpenAI says - OpenAI states that creating AI tools like ChatGPT without copyrighted material is impossible, as pressure mounts on AI firms over the content used to train their products.
OpenAI warns copyright crackdown could doom ChatGPT - OpenAI warns that a ban on using news and books to train chatbots could doom the development of artificial intelligence, as it seeks to influence potential laws on the topic and faces lawsuits from book publishers and the New York Times.
AI-Generated George Carlin Comedy Special Slammed by Comedian’s Daughter - AI-generated George Carlin comedy special sparks outrage from comedian's daughter, who criticizes the attempt to recreate her father's genius and suggests listening to the genuine Carlin instead.
No, That’s Not Taylor Swift Peddling Le Creuset Cookware - AI technology has been used to create deceptive ads featuring synthetic versions of celebrities, including Taylor Swift, promoting products without their endorsement.
OpenAI Quietly Deletes Ban on Using ChatGPT for “Military and Warfare” - OpenAI quietly removes ban on using ChatGPT for military and warfare, raising concerns about its potential use by the military despite ethical and safety implications.
The New York Times’ AI Opportunity - The article discusses the legal battle between The New York Times and OpenAI over copyright infringement, focusing on the use of AI to train chatbots using copyrighted material, and the implications for fair use and the value of authoritative content creators in the digital age.
Judges in England and Wales are given cautious approval to use AI in writing legal opinions - Judges in England and Wales are cautiously allowed to use AI to write legal opinions, with the guidance emphasizing the need for personal responsibility, caution about the limitations and potential biases of AI, and the importance of keeping humans in the loop.
UK government to publish ‘tests’ on whether to pass new AI laws - UK government to publish ‘tests’ on whether to pass new AI laws, as part of efforts to regulate the use of artificial intelligence in the country.
Valve opens the door to more Steam games developed with AI - Valve introduces new rules for AI-powered games on Steam, requiring developers to disclose AI usage and ensure it does not generate illegal content, aiming to increase transparency and protect against potential risks.
California AG Must Investigate OpenAI’s Non-Profit Status - Public Citizen calls on California AG to investigate whether OpenAI should retain its non-profit status, citing concerns that the organization may be prioritizing profit over its non-profit purpose.
Copyright © 2024 Skynet Today, All rights reserved.