Last Week in AI #252: DeepMind's real-world robots, NYT sues OpenAI, Baidu's Ernie bot hits 100M users, AI art generator controversy, and more! π€π°ππ¨
DeepMind is trying to train robots for real-world activities, The Times Sues OpenAI and Microsoft Over A.I. Use of Copyrighted Work, Baidu says its Ernie bot now has more than 100 million users
Top News
DeepMind is trying to train robots for real-world activities
Google's DeepMind is developing multiple research projects aimed at creating robots capable of making faster decisions and functioning in real-world scenarios. The first project, AutoRT, combines large foundational models with a robot control model, enabling robots to gather training data in new environments and multitask. The system has been tested in real-world settings, directing up to 20 robots simultaneously in various tasks. To ensure safety, DeepMind has integrated a "robot constitution" into AutoRT, which includes safety measures from classical robotics and Isaac Asimovβs Three Laws of Robotics. Another system, SARA-RT, improves the efficiency of robotic transformer models, while the third project, RT-Trajectory, helps robots become more generalized in their tasks by providing visual hints during training. DeepMind's efforts are part of a broader trend towards integrating AI and robotics in practical, everyday applications.
The Times Sues OpenAI and Microsoft Over A.I. Use of Copyrighted Work
The New York Times has filed a lawsuit against OpenAI and Microsoft, accusing them of copyright infringement for using its published work to train artificial intelligence technologies without authorization. The lawsuit, the first of its kind from a major American media organization, alleges that millions of Times articles were used to train chatbots, which now compete with the Times as a source of information. The Times is seeking "billions of dollars in statutory and actual damages" and demands the destruction of any chatbot models and training data that use its copyrighted material. The Times had previously approached both companies in April to discuss its concerns and seek a resolution, but no agreement was reached.
Baidu says its ChatGPT rival Ernie bot now has more than 100 million users
Chinese tech giant Baidu has announced that its artificial intelligence (AI) chatbot, Ernie bot, has reached over 100 million users. This AI product, similar to Microsoft-backed OpenAI's ChatGPT, is primarily used in Chinese but also supports English. To register, users need a China mobile number. The chatbot, known as "Wenxinyiyan" in Mandarin, was launched in March but only received regulatory approval for widespread use in late August. The company did not clarify whether the reported user numbers were active or for a specific time period.
A list going viral reveals famous artists whose work was used to train AI generator
A list revealing thousands of artists whose work was used to train an artificial intelligence (AI) art generator has gone viral, sparking a lawsuit against companies Midjourney, Stability AI, DeviantArt, and Runway AI. The companies are accused of misusing copyrighted work from visual artists to train their generative AI systems, with many artists alleging that their work was stolen without permission. The list, which includes nearly 4,700 artists and an additional 16,000 proposed additions, has highlighted frustrations with the lack of regulation around AI-generated art. The lawsuit argues that the AI models are built around human intelligence and creative expression, and that the profits from the misappropriation of these works flow directly into the defendants' pockets without consent, credit, or compensation to the artists.
Other News
Tools
Microsoft quietly launches dedicated Copilot app for Android - Microsoft quietly launched a new Copilot app on Android, powered by AI, allowing users to interact with advanced OpenAI models for chat and visual creation.
Microsoftβs next Surface laptops will reportedly be its first true βAI PCsβ - Microsoft's upcoming Surface Pro 10 and Surface Laptop 6 will be the company's first true 'AI PCs,' featuring new AI-enabled features and next-gen neural processing units, along with improved performance, battery life, and security.
Android Auto will have Google Assistant summarize your messages with AI - Google is developing a feature for Android Auto that will use Google Assistant to summarize messages and busy conversations using AI, which can be turned on or off in the settings.
Some of the Samsung Galaxy S24's key AI features just leaked - Samsung Galaxy S24's key AI features include live translation, 'nightography' zoom, and generative edit, potentially exclusive to the more expensive models, aiming to keep up with Google's AI-packed Pixel phones.
GitHub Copilot Chat now generally available for organizations and individuals - GitHub Copilot Chat, powered by GPT-4, is now generally available for individuals and organizations, providing real-time coding assistance in multiple natural languages and seamless translation between programming languages.
Open source AI voice cloning arrives with MyShellβs new OpenVoice model - OpenVoice, an open-source voice cloning model developed by researchers at MIT and MyShell, offers granular control over tone, emotion, and accent, allowing users to create voice clones with just a small audio clip.
These AI-powered apps can hear the cause of a cough - AI-powered apps using cough or speech patterns to alert to health problems are on the rise, with potential for diagnosing various conditions, but concerns about real-world performance and the need to balance AI with clinical judgment remain.
Microsoft Adds AI Key in First Change to PC Keyboard in Decades - Microsoft introduces a new PC keyboard with an AI key, marking the first significant change to the keyboard in decades.
Business
OpenAI Is in Talks to Raise New Funding at Valuation of $100 Billion or More - OpenAI is in discussions to raise new funding at a valuation of $100 billion or more, potentially making it one of the worldβs most valuable startups, as it continues to attract significant investment interest and expand its AI capabilities.
Shield AI Raises Additional $100M in Series F; $200M in Debt - Shield AI secures $100M in Series F funding and $200M in debt to accelerate deployment of AI pilots for various aircraft, including its flagship product, Hivemind, an AI pilot for autonomous missions in high-threat environments.
AIβs big test: Making sense of $4 trillion in medical expenses - AI is being adopted by hospitals and insurers to process medical bills, with the potential to increase revenue, reduce administrative workforces, and streamline the prior-authorization process, but policymakers are just beginning to grapple with its implications.
AI is saving sales professionals more than two hours of work each day - AI is saving sales professionals over two hours of work each day by automating tasks like scheduling meetings and note-taking, allowing them to focus more on connecting with customers and closing deals.
AI-powered search engine Perplexity AI, now valued at $520M, raises $70M - Perplexity AI, a startup search engine, raises $70M in funding, aiming to reinvent AI-powered search with a chatbot-like interface and GenAI models, despite concerns about cost, misuse, and copyright issues.
AI-powered search engine Perplexity AI, now valued at $520M, raises $73.6M - Perplexity AI, a startup search engine, has raised $73.6 million in funding and is valued at $520 million, offering a chatbot-like interface with AI-powered search capabilities and plans to expand its GenAI models and features.
Nvidia releases slower, less powerful AI chip for China - Nvidia has released a new gaming processor in China that complies with US export rules, offering a scaled-back version of its RTX 4090 chip with 11% slower processing rate to maintain its dominance in the Chinese market amidst US bans on high-tech exports.
Aurora finalizes design of self-driving trucks it will make with Continental - Aurora finalizes design, architecture, and hardware for its self-driving trucks in partnership with Continental, aiming to deploy "thousands" of trucks by 2027 and taking a measured, conservative approach to commercialization.
Microsoft is Rebranding Edge on Mobile to Microsoft Edge: AI Browser, Starting a New Era of AI Services - Microsoft rebrands its mobile browser to Microsoft Edge: AI Browser, highlighting its AI capabilities in response to the AI race of 2024.
Microsoft, OpenAI sued for copyright infringement by nonfiction book authors in class action claim - Authors sue Microsoft and OpenAI for copyright infringement, alleging that the companies used their copyrighted works to help build a billion-dollar artificial intelligence system, seeking damages for a broader class of plaintiffs.
Microsoft Picks Dee Templeton as OpenAI Board Observer - Microsoft appoints Dee Templeton as an observer on the OpenAI board.
Research
New research harnesses AI and satellite imagery to reveal the expanding footprint of human activity at sea - AI and satellite imagery reveal previously unmapped industrial use of the ocean, detecting hidden vessel activity and offshore infrastructure, including the revelation that 75 percent of the worldβs industrial fishing vessels are not publicly tracked.
Unified-IO 2: Scaling Autoregressive Multimodal Models with Vision, Language, Audio, and Action - A large multimodal model, Unified-IO 2, with 7 billion parameters is presented, capable of encoding and producing text, image, audio, video, and sequences, and setting new benchmarks across various modalities.
SOLAR 10.7B: Scaling Large Language Models with Simple yet Effective Depth Up-Scaling - Scaling large language models with a simple and effective depth up-scaling method, DUS, allows for the creation of SOLAR 10.7B, an LLM with 10.7 billion parameters that outperforms existing models in various benchmarks.
Improving Text Embeddings with Large Language Models - Improving text embeddings with large language models through a novel method leveraging LLMs to generate synthetic data for diverse text embedding tasks in multiple languages, achieving competitive performance and state-of-the-art results.
Video Understanding with Large Language Models: A Survey - Advancements in large language models have revolutionized video understanding, enabling them to process and interpret complex interactions between visual and textual data, and offering a comprehensive survey on the topic.
PanGu-$Ο$: Enhancing Language Model Architectures via Nonlinearity Compensation - PanGu-Ο introduces a new architecture for language models, addressing the feature collapse problem via nonlinearity compensation, achieving better performance and efficiency in NLP tasks.
TinyGPT-V: Efficient Multimodal Large Language Model via Small Backbones - A new model called TinyGPT-V is proposed, utilizing an advanced large language model and pre-trained vision modules to rival the performance of larger models while requiring significantly fewer computational resources.
Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models - Self-Play Fine-Tuning (SPIN) is a novel method that empowers weak language models to improve themselves through self-play, eliminating the need for additional human-annotated data and achieving significant performance improvements.
DocLLM: A layout-aware generative language model for multimodal document understanding - A new layout-aware generative language model, DocLLM, is introduced to address the challenges of understanding visually rich documents, incorporating spatial layout information and achieving performance improvements in various document intelligence tasks.
Researchers use AI chatbots against themselves to 'jailbreak' each other - AI chatbots have been compromised by researchers using a method called "jailbreaking," which involves training a large language model to automatically generate prompts that bypass the chatbots' ethical guidelines, leading to the production of restricted information.
LARP: Language-Agent Role Play for Open-World Games - AI language agents are being integrated into open-world games through a framework called LARP, which focuses on creating a realistic role-playing experience by blending cognitive architecture, memory processing, and decision-making.
Mobile ALOHA: Learning Bimanual Mobile Manipulation with Low-Cost Whole-Body Teleoperation - Imitation learning from human demonstrations is used to develop a system for learning complex mobile bimanual manipulation tasks, with the development of a low-cost whole-body teleoperation system and the finding that co-training enables data-efficient learning of complex mobile manipulation tasks.
A Comprehensive Survey of Hallucination Mitigation Techniques in Large Language Models - Mitigating hallucination in large language models is a critical focus in contemporary computational linguistics, with researchers proposing various strategies to address this challenge, leading to the consolidation and organization of diverse techniques into a comprehensive taxonomy.
Task Contamination: Language Models May Not Be Few-Shot Anymore - AI language models may not be as few-shot as previously thought, as task contamination from pre-training data can significantly impact their performance across various tasks and models.
Gemini in Reasoning: Unveiling Commonsense in Multimodal Large Language Models - Gemini Pro's performance in commonsense reasoning tasks is evaluated across 12 diverse datasets, showing comparable results to GPT-3.5 Turbo in language-based tasks but lagging behind in accuracy and facing challenges in temporal, social reasoning, and emotion recognition in images.
GPT-4V(ision) is a Generalist Web Agent, if Grounded - GPT-4V(ision) is a strong generalist web agent, demonstrating potential for web agents but facing challenges with grounding strategies.
FlowVid: Taming Imperfect Optical Flows for Consistent Video-to-Video Synthesis - A new V2V synthesis method called FlowVid harnesses optical flow benefits and handles imperfections, achieving efficiency and high generation quality.
DeWave: Discrete EEG Waves Encoding for Brain Dynamics to Text Translation - Decoding brain states into comprehensible representations through a pioneering framework called DeWave, which uses discrete codex encoding to translate EEG waves into text, achieving state-of-the-art performance in EEG translation.
Google wrote a βRobot Constitutionβ to make sure its new AI droids wonβt kill us - Google's DeepMind robotics team has developed new advances, including a "Robot Constitution," to ensure robots can make faster, better, and safer decisions, with features such as a data gathering system, safety prompts, and physical kill switches.
LLaMA Pro: Progressive LLaMA with Block Expansion - LLAMA PRO introduces a block expansion method to enhance large language models' domain-specific abilities while preserving their general capabilities, leading to the development of a versatile and powerful model that excels in general, mathematical, and programming tasks.
Meet LLM Surgeon: A New Machine Learning Framework for Unstructured, Semi-Structured, and Structured Pruning of Large Language Models (LLMs) - Advancements in AI have led to the development of Large Language Models (LLMs) with billions of parameters, and the LLM Surgeon framework effectively prunes these models by up to 30% without significant performance loss, addressing the challenge of deploying such large models.
Concerns
A New Kind of AI Copy Can Fully Replicate Famous People. The Law Is Powerless. - AI technology has advanced to the point where it can replicate famous individuals, raising ethical and legal concerns about consent and intellectual property rights.
Microsoft says its AI is safe. So why does it keep slashing peopleβs throats? - Microsoft's AI Image Creator is making extremely violent and disturbing images, and despite Microsoft's claims of safety, the company's response to the issue has been inadequate.
AI altered a Keith Haring painting about the AIDS crisis β and, for some, ruined its meaning - AI was used to "complete" Keith Haring's "Unfinished Painting" about the AIDS crisis, sparking outrage among artists who argue that using AI destroyed the pieceβs meaning and raised concerns about the perils of creating images with AI using other peopleβs original work.
Former Trump lawyer Michael Cohen accidentally cited fake court cases generated by AI - Michael Cohen admitted to citing fake, AI-generated court cases in a legal document, mistakenly using Google's Bard as a search engine, which led to potential sanctions for his lawyer.
Policy
AIβs future could hinge on one thorny legal question - AI's future could hinge on the legal question of whether tech firms' use of copyrighted works to train AI models constitutes fair use or infringement, with potential consequences for the booming generative AI industry and media companies.
How the Federal Government Can Rein In A.I. in Law Enforcement - Federal Office of Management and Budget proposes guidelines for law enforcement's use of AI technologies, aiming to increase transparency, assess risks, and prevent biases and errors.
California senator files bill prohibiting agencies from working with unethical AI companies - California senator introduces bills to regulate AI systems used by state agencies, establishing safety, privacy, and non-discrimination standards, and creating a public AI resource and research hub.
Whatβs next for AI regulation in 2024?Β - AI regulation in 2024 will bring new rules and standards for high-risk AI applications in the EU, requiring transparency, accountability, and compliance with EU standards.
FTC offers $25,000 prize for detecting AI-enabled voice cloning - FTC is offering a $25,000 prize for ideas to protect consumers from the danger of AI-enabled voice cloning for fraudulent activity, as voice cloning technology poses a significant risk for fraudulent scams and deception.
When driverless cars speed or run red lights in San Francisco, they can't be ticketed - Driverless cars in San Francisco cannot be ticketed for moving violations, as local police lack the authority to cite the empty vehicles, prompting discussions about potential legislative changes.
Analysis
2023: The Year of AI - AI has made significant advancements in 2023, with notable progress in image and video generation, text-to-image algorithms, and partnerships, while also facing legal and ethical debates and challenges.
Robotics in the Era of Foundation Models - 2023 saw a significant surge in AI and robotics, with a focus on foundation models, generalist robot models, data engines, low-level control, and the emergence of humanoid robotics startups.
The Hollywood Strikes Stopped AI From Taking Your Job. But for How Long? - The Hollywood strikes and unions' resistance against AI in 2023 set a precedent for future labor movements to push back against encroaching automation, as various professions found themselves vulnerable to being replaced by machine learning.
A case for AI alignment being difficult - AI alignment is a complex challenge, involving the definition of human values, normative criteria for AI, consequentialism, problem-solving methods, and the difficulty of specifying optimization of a different agent's utility function, with potential paths forward including human enhancement and simulated humans.
Fun
Happy Puppies and Silly Geese: Pushing the Limits of A.I. Absurdity - AI is being used to create absurd and vibrant images, pushing the boundaries of its capabilities and delighting social media users.
Copyright Β© 2024 Skynet Today, All rights reserved.