Last Week in AI #247: Sam Altman returns to OpenAI, Cruise CEO resigns, Stability releases video diffusion models, Underage workers are labeling AI data, and more!
Dust settles at OpenAI after 5 days of chaos, Cruise CEO resigns amidst safety investigations, open source video models from Stability AI, underage workers found to be labeling AI training data
Top News
Five Days of Chaos: How Sam Altman Returned to OpenAI
Sam Altman has returned as CEO of OpenAI just five days after being ousted by the company's board of directors. His reinstatement followed an intense pressure campaign and negotiations, marking the end of a chaotic period for the AI start-up. The company will now replace its board with a more traditional group, including former Treasury Secretary Lawrence Summers and a former Salesforce executive. Despite the controversy, some believe this recent upheaval could result in a more valuable and better-equipped organization. The company's future board members, potentially from its largest investor, Microsoft, and the AI research community, are expected to be announced soon. These changes come amidst concerns that OpenAI's initial mission to create safe AI could be compromised due to the increasing commercial focus.
Cruise co-founder and CEO Kyle Vogt resigns
Kyle Vogt, the co-founder and CEO of self-driving car company Cruise, has stepped down from his role, as has co-founder Dan Kan. The resignations follow a recent incident where a Cruise robotaxi hit a pedestrian, leading the California Department of Motor Vehicles to suspend the company's permits to operate autonomous vehicles on public roads. In the wake of the incident, the company laid off contract workers and morale among employees has reportedly been low. General Motors (GM) Chair and CEO Mary Barra announced that Mo Elshenawy, currently executive vice president of engineering at Cruise, will serve as president and CTO for the company. No new CEO has been named yet. Vogt and Kan founded the autonomous vehicle company in 2013, which was acquired by GM in 2016.
Stability AI debuts Stable Video Diffusion models in research preview
Stability AI has released Stable Video Diffusion (SVD), a set of AI models for generating short video clips from still images. The models, SVD and SVD-XT, produce high-quality outputs and have been open-sourced for research purposes. The models take in a still image as a conditioning frame and generate 576 x 1024 videos, lasting up to four seconds. Stability AI plans to refine the models based on user feedback and eventually introduce new features for commercial applications. The models have been evaluated to have high quality outputs but still have limitations, such as missing photorealism and generating videos without motion or faces. The code and weights for the models are available on Stability AI's GitHub repository and Hugging Face page.
Underage Workers Are Training AI
Big tech companies are indirectly employing underage workers to label data for AI systems, often exposing these young teens to traumatic content. The raw data used in machine learning algorithms is labeled by humans, a task often outsourced to gig workers via platforms like Toloka. These workers, some under 18, classify and moderate content from the internet, including harmful and explicit material. The global data labeling industry is projected to reach over $17.1 billion by 2030. Crowdsourcing platforms like Toloka, Appen, and Clickworker connect remote gig workers from countries like Pakistan, India, the Philippines, Venezuela, and East Africa to tech companies in Silicon Valley. Despite platforms requiring workers to be over 18, many underage individuals bypass these checks by using a relative's details and corresponding payment methods. The labor these workers provide is considered essential for AI development, yet their pay is significantly lower than in-house employees of tech companies, leading to criticism of digital colonialism.
Other News
Tools
Microsoft releases AI tool for photorealistic copying of faces and voices - Microsoft has released a new AI tool called Azure AI Speech that can generate photorealistic avatars and replicate voices, raising concerns about the creation of deepfakes.
Be My Eyes AI offers GPT-4-powered support for blind Microsoft customers - Microsoft is partnering with Be My Eyes to integrate OpenAI's ChatGPT-4 into its customer service for visually impaired users, allowing them to resolve technical issues and perform tasks without the need for human assistance.
Meta launches AI-based video editing tools - Meta Platforms has launched two new AI-based video editing tools, Emu Video and Emu Edit, which allow users to generate short videos with captions and easily alter or edit videos with text prompts, as the company continues to make advancements in the AI space to compete with other tech giants.
Google will make fake AI products to help you find real gifts - Google's AI-powered Search Generative Experience (SGE) is introducing new shopping features that use AI to generate gift ideas and fashion items, as well as allowing users to generate photorealistic images of apparel they want and find similar-looking real products to purchase online.
Spotify to use Google's AI to tailor podcasts, audiobooks recommendations - Spotify is expanding its partnership with Google Cloud to use large language models (LLMs) in order to tailor podcast and audiobook recommendations based on a user's listening patterns.
Runway’s New ‘Motion Brush’ Feature in Gen-2 will Allow to Add Controlled Movement to Your Generations - Runway's Gen-2 introduces the Motion Brush feature, allowing users to effortlessly create videos without complex text inputs or extensive editing, revolutionizing video generation and paving the way for a more intuitive and efficient future in video creation.
OpenAI rival Anthropic makes its Claude chatbot even more useful - Anthropic's chatbot, Claude 2.1, has received an update that allows it to process up to 200,000 tokens at once, making it more powerful than OpenAI's GPT-4 and enabling it to analyze large amounts of text or code.
Screenshots show xAI’s chatbot Grok on X’s web app - Screenshots reveal the development of xAI's chatbot Grok, which will be part of the top-tier subscription service X Premium+, offering a conversational and witty AI assistant with access to real-time knowledge and the ability to answer "spicy" questions.
Formula 1 hopes AI will help it figure out if a car breaks track limits - Formula 1 will test an AI system using Computer Vision tech to determine if cars go out of bounds during races, aiming to reduce the number of potential rule violations that need manual review by officials.
Discord is already killing Clyde, its experimental OpenAI chatbot - Discord is shutting down its AI chatbot Clyde, less than a year after its introduction, due to reasons that are not specified in the article.
AI cleaning robot can tidy up clothes in a messy bedroom - A robot with cameras and a grasping arm has been developed to efficiently pick up and tidy piles of clothes in messy bedrooms.
What was the GAN? - Generative Adversarial Networks (GANs) were a popular tool for AI artists, allowing them to generate photorealistic images based on patterns found in large datasets, but now RunwayML, a web-based app and AI playground for creatives, has announced that it will be discontinuing its machine learning tools, including GANs, in favor of diffusion-based images and video.
Business
Amazon lays off hundreds in its Alexa division as it plows resources into AI - Amazon is cutting hundreds of jobs in its Alexa division as it focuses more on artificial intelligence, specifically generative AI.
A.I. Belongs to the Capitalists Now - The battle between two dueling visions of artificial intelligence, one seeing it as a transformative tool and the other as a potentially dangerous force, appears to be over with the return of Sam Altman to OpenAI.
Exclusive: OpenAI investors considering suing the board after CEO's abrupt firing - sources - Investors in OpenAI are considering legal action against the company's board after the abrupt firing of CEO Sam Altman, which has led to concerns about potential financial losses and a mass exodus of employees.
Meta disbanded its Responsible AI team - Meta has disbanded its Responsible AI team and shifted its focus to generative artificial intelligence, with most team members moving to the generative AI product team and others working on Meta's AI infrastructure.
Amazon aims to provide free AI skills training to 2 million people by 2025 with its new ‘AI Ready’ commitment - Amazon is launching "AI Ready," a commitment to provide free AI skills training to 2 million people by 2025 through new initiatives and collaborations, including free courses, scholarships, and an Hour of Code Dance Party: AI Edition.
Airbnb acquires AI startup for just under $200 million - Airbnb acquires AI startup Gameplanner.AI for just under $200 million, which will accelerate some of Airbnb's AI projects.
NVIDIA Announces Financial Results for Third Quarter Fiscal 2024 - NVIDIA reported record revenue of $18.12 billion for the third quarter of fiscal 2024, up 34% from the previous quarter and up 206% from a year ago, driven by growth in data center revenue and the transition to accelerated computing and generative AI.
AI21 takes Series C to $208 million with additional $53 million in funding - AI21, a company that develops AI systems for enterprise, has completed its $208 million Series C funding round, bringing its total capital raised to $336 million, and plans to recruit an additional 100 employees in 2024.
Kyutai is a French AI research lab with a $330 million budget that will make everything open source - French billionaire Xavier Niel has revealed plans for Kyutai, a privately-funded nonprofit AI research lab in Paris, with a budget of €300 million ($330 million), which will focus on artificial general intelligence and make its research papers and open source projects available to the public.
Off/Script launches an app to create and buy AI-designed fashion - Off/Script has launched a mobile app that allows users to create and monetize AI-designed product mock-ups, with the most popular designs being funded, manufactured, and shipped by the Montreal-based startup.
GM Cruise cofounder, senior exec Dan Kan quits day after CEO exit - GM Cruise co-founder and chief product officer Daniel Kan has resigned, a day after Cruise CEO Kyle Vogt quit, amid a safety review of the company's self-driving taxi fleet.
US chip export ban is hurting China’s AI startups, not so much the giants yet - The US chip export ban is hurting China's AI startups, while the tech giants have been hoarding chips in anticipation of a tech war between the two nations.
C3.ai cuts jobs to reduce costs - Bloomberg News - Software company C3.ai has cut jobs across departments, citing employee performance and the need for cost savings.
A Spanish agency became so sick of models and influencers that they created their own with AI—and she’s raking in up to $11,000 a month - A Spanish agency created an AI model named Aitana to avoid the costs and difficulties associated with working with real-life models and influencers, and she has become a successful influencer herself, earning up to $11,000 a month.
OpenAI's board approached Anthropic CEO about top job and merger - sources - OpenAI's board approached Anthropic's CEO about replacing their chief and potentially merging the two AI startups, but the CEO declined both offers.
Research
Meet JARVIS-1: Open-World Multi-Task Agents with Memory-Augmented Multimodal Language Models - Researchers have developed JARVIS-1, a multimodal agent designed for open-world tasks in Minecraft, which leverages pre-trained multimodal language models to interpret visual observations and human instructions, generating sophisticated plans for embodied control.
Synthetic imagery sets new bar in AI training efficiency - MIT researchers have developed a system called StableRep that uses synthetic images generated through text-to-image models to train machine learning models, surpassing the results obtained from traditional "real-image" training methods.
Video-LLaVA: Learning United Visual Representation by Alignment Before Projection - A new approach called Video-LLaVA has been developed to unify visual representation and language feature space, resulting in a robust large vision-language model that outperforms existing models on various image benchmarks and video datasets.
UFOGen: You Forward Once Large Scale Text-to-Image Generation via Diffusion GANs - A new AI model called UFOGen uses Diffusion GANs to generate large-scale text-to-image conversions, allowing users to upload images, audio, and videos through various methods.
ShareGPT4V: Improving Large Multi-Modal Models with Better Captions - Introducing the ShareGPT4V dataset, a large-scale resource of highly descriptive captions that significantly enhances large multi-modal models, improving their performance on various benchmarks.
The Chosen One: Consistent Characters in Text-to-Image Diffusion Models - The article discusses the concept of consistent characters in text-to-image diffusion models and explores the idea of using a magic wand tool to exclude certain pixels from being manipulated in subsequent generations, allowing for dynamic yet consistent design elements.
Chain-of-Note: Enhancing Robustness in Retrieval-Augmented Language Models - Chain-of-Noting (CoN) is a novel approach that improves the robustness of retrieval-augmented language models (RALMs) by generating sequential reading notes for retrieved documents, resulting in better performance in handling noisy and irrelevant data as well as unknown scenarios.
Diffusion Model Alignment Using Direct Preference Optimization - Diffusion-DPO is a method proposed to align diffusion models to human preferences by directly optimizing on human comparison data, improving visual appeal and prompt alignment in text-to-image diffusion models.
Generative AI Passes the Legal Ethics Exam in Study by LegalOn Technologies - Generative AI models, including OpenAI's GPT-4 and Anthropic's Claude 2, have been shown to pass the legal ethics exam, demonstrating their potential to assist lawyers in legal work and highlighting the advanced capabilities of large language models applied to law.
Orca 2: Teaching Small Language Models How to Reason - The article discusses the development of Orca 2, a system that teaches small language models how to reason.
GAIA: a benchmark for General AI Assistants - The article discusses the potential lack of representativeness in the educational background of annotators in a benchmark for General AI Assistants, raising questions about the research sample's alignment with the general population.
LLMs cannot find reasoning errors, but can correct them! - LLMs have the ability to correct reasoning errors, a feature that has been embraced by individuals and organizations working with arXivLabs.
System 2 Attention (is something you might need too) - The Librarian Bot found similar papers and recommends them using the Semantic Scholar API.
Exponentially Faster Language Modelling - The article discusses slower inference times for a language model compared to a standard model, and highlights the need to consider the code and setup used for benchmarking performance speedups.
Will Scaling Solve Robotics?: Perspectives From Corl 2023 - The article discusses the debate surrounding whether training large neural networks on large datasets is a feasible way to solve robotics, exploring arguments for and against scaling as a solution, as well as related points such as the need for human-in-the-loop systems and the potential of combining classical and learning-based approaches.
Contrastive Chain-of-Thought Prompting - The article discusses the process of uploading various types of media, such as images, audio, and videos, through different methods.
Concerns
OpenAI's alignment problem - OpenAI's board fired CEO Sam Altman due to concerns that the company was unduly accelerating the development of artificial general intelligence (AGI), leading to a breakdown in communication between Altman and the board.
OpenAI and Microsoft sued over alleged unauthorized use of authors’ work - OpenAI and Microsoft are being sued by author Julian Sancton for allegedly using nonfiction authors' work without authorization to train AI models, including OpenAI's ChatGPT service.
Microsoft AI Image Generator Blocks ‘Disney’ After Viral Movie Poster Trend - Microsoft has blocked the term "Disney" from its AI image generator tool after a viral trend of AI-generated movie posters featuring dogs as main characters raised concerns over copyright infringement.
Screensharing mistake gets Nvidia sued over alleged stolen trade secrets - Nvidia is being sued by automotive technology company Valeo for allegedly benefiting from stolen trade secrets after an employee mistakenly shared source code during a Microsoft Teams call.
Several popular AI products flagged as unsafe for kids by Common Sense Media - Popular AI products, including Snapchat's My AI, DALLE, and Stable Diffusion, have been flagged as unsafe for kids by Common Sense Media due to concerns about biases, privacy, and inappropriate responses.
‘Lost Time for No Reason’: How Driverless Taxis Are Stressing Cities - Driverless taxis operated by Waymo, an autonomous car company owned by Alphabet, are causing delays in emergency response times in San Francisco due to their inability to be instructed by a human driver to move out of the way.
Sarah Silverman Hits Stumbling Block In AI Copyright Infringement Lawsuit Against Meta - A federal judge has dismissed most of Sarah Silverman's lawsuit against Meta over the unauthorized use of authors' copyrighted books to train its generative artificial intelligence model, marking the second ruling from a court siding with AI firms on novel intellectual property questions presented in the legal battle.
AI Art Generators Can Be Fooled Into Making NSFW Images - Nonsense words can deceive popular AI art generators like DALL-E 2 and Midjourney into generating explicit and controversial images.
More than half of Americans are worried about AI than excited - More than half of Americans are concerned about the role AI may play in their lives, with fears of harm to human relationships, job loss, and privacy being the main areas of concern, although there is also optimism about AI's potential to remove bias in healthcare and hiring.
Analysis
Why it’s important to remember that AI isn’t human - Language models like ChatGPT should not be thought of as human-like beings, as they lack the same psychological properties and motivations, and thinking of them as such can impede scientific progress and appropriate AI regulation.
Making Sense of OpenAI, Sam Altman and the Non-profit Board on Their Hardest Days - The recent events at OpenAI, including the firing and rehiring of CEO Sam Altman, have exposed the inherent incoherences and implausible aspirations of the company, which should abandon its divine aura, drop the "for all of humanity" narrative, and become more candid and transparent in its communication with the world.
Policy
France, Germany and Italy join forces to propose unified AI regulation in the EU - France, Germany, and Italy propose a unified AI regulation in the EU that advocates for mandatory self-regulation and voluntary rules for AI providers without initial sanctions.
OpenAI’s Misalignment and Microsoft’s Gain - OpenAI's CEO, Sam Altman, was fired by the board and subsequently joined Microsoft, leading to concerns about the misalignment of OpenAI's non-profit model and its mission to develop safe AI technology.
Before Altman’s Ouster, OpenAI’s Board Was Divided and Feuding - Tensions and feuds within OpenAI's board of directors, including the ousting of Sam Altman and the departure of critical members, have highlighted the challenges of aligning business interests with the concerns of researchers regarding the potential impact of AI on jobs and the development of autonomous weapons.
Music publishers ask court to halt AI company Anthropic's use of lyrics - Three music publishers are seeking a preliminary injunction to prevent AI company Anthropic from reproducing or distributing their copyrighted song lyrics, accusing the company of "systematic and widespread" infringement.
Canadian Chamber of Commerce forms AI council with members including Amazon, Google - The Canadian Chamber of Commerce has formed an AI council, including members such as Amazon and Google, to advocate for government policies centered on responsible development, deployment, and ethical use of AI in business.
State Dept prioritizes ‘AI-ready workforce’ in its first AI strategy - The State Department has released its first enterprise AI strategy, prioritizing the development of an "AI-ready workforce" to maximize the impact of its diplomats and turn data into real-world diplomatic insights.
Inside the Chaos at OpenAI - OpenAI's recent turmoil, including the ousting of CEO Sam Altman and his subsequent decision to start a new AI team at Microsoft, highlights the tension between the company's mission to create beneficial artificial general intelligence and the pressure to commercialize its products.
Businesses, tech groups warn EU against over-regulating AI foundation models - Businesses and tech groups are warning the European Union against over-regulating AI foundation models in upcoming AI rules, as this could harm start-ups and drive them out of the region.
As A.I.-Controlled Killer Drones Become Reality, Nations Debate Limits - Nations are debating whether to adopt nonbinding guidelines or substantive legally binding restrictions on the use of lethal autonomous weapons, as concerns over the risks of artificial intelligence and the potential for A.I.-controlled killer drones increase.
UK will refrain from regulating AI ‘in the short term’ - The UK will not regulate AI in the short term, according to the article.
Copyright © 2023 Skynet Today, All rights reserved.