- AI Breakfast
- Posts
- NYT lawsuit could force OpenAI to wipe ChatGPT and start over
NYT lawsuit could force OpenAI to wipe ChatGPT and start over
AI Breakfast
August 18, 2023
Good morning. It’s Friday, August 18th.
Did you know: Robotics company Somatic has developed a janitorial robot that can clean toilets by itself?
In today’s email:
AI in Software & Tools Development
Legal & Ethical Concerns in AI
AI Business & Market Trends
AI in Entertainment & Content Generation
AI in Research & Science
5 New AI Tools
Latest AI Research Papers
You read. We listen. Let us know what you think of this edition by replying to this email, or DM us on Twitter.
Today’s edition is brought to you by:
Raindrop helps you design personalized trading strategies using AI.
Whether you're a beginner or an experienced trader, Raindrop helps you write custom trading bots tailored to your investment strategy.
Simplify and automate your portfolio management with Raindrop’s AI Bot Marketplace.
Can you program an intelligent trading bot of your own?
Today’s trending AI news stories
AI in Software & Tools Development
Meta’s Next AI Attack on OpenAI: Free Code-Generating Software Meta Platform is set to launch open-source code-generating AI software, Code Llama, aimed at helping developers automatically generate programming code. This move is seen as a challenge to proprietary offerings from OpenAI, Google, and others. The software, an extension of Meta’s Llama 2 large-language model, enables AI assistants to suggest code as developers type. It may compete with paid coding assistants like GitHub Copilot. Meta’s approach follows its previous success with Llama 2 in democratizing app development. The software could potentially disrupt the coding assistance market and impact offerings from rivals like OpenAI and Microsoft.
Arthur unveils Bench, an open-source AI model evaluator: Arthur, an AI startup based in New York City, has introduced Arthur Bench, an open-source tool designed for evaluating and comparing the performance of LLMs like OpenAI’s GPT 3.5 Turbo and Meta’s LLaMA 2. The tool enables companies to test LLMs on specific use cases, comparing metrics such as accuracy, readability, and hedging. Arthur Bench accelerates benchmarking by translating academic measures into real-world business impact. Financial services firms and vehicle manufacturers are already using the tool, and Arthur is collaborating with Amazon Web Services and Cohere for further development.
Alibaba Cloud launches AI digital human video generation tool to expand image-generated video applications: Alibaba Cloud’s subsidiary, Aliyun, has unveiled Live Portrait, an AI-powered digital human video generation tool. By uploading a photo and text or voice, users can create talking digital human videos for applications like live video streaming, chatbots, and corporate marketing. The tool’s innovative approach breaks traditional generation network limitations, offering broad applications and potential for cost-efficient enterprise production. Live Portrait currently offers text and audio-driven modes with various language options, aiming to enhance efficiency and engagement across industries.
Adobe Express with AI Firefly app is available worldwide: Adobe’s AI-powered image editing tool, Adobe Express with Firefly, has exited beta and is now available for free use in web browsers. The tool, aimed at simplifying content creation, offers generative AI prompts in over 100 languages, making it easier to generate high-quality images. The web app, known as Adobe Spark until 2021, does not require a subscription to Adobe’s Creative Cloud Service for basic use, but a premium version is available for subscribers. The mobile version of Adobe Express with Firefly is expected to be released soon.
Legal & Ethical Concerns in AI
Report: Potential NYT lawsuit could force OpenAI to wipe ChatGPT and start over: The New York Times is reportedly considering a lawsuit against OpenAI, which could lead to devastating consequences for OpenAI, including the potential destruction of ChatGPT’s dataset and fines up to $150,000 per infringing content piece. The Times’ updated terms of service (TOS) prohibit AI companies from scraping their content for training AI models. If the lawsuit proceeds, it could become a high-profile legal battle over copyright protection and AI’s use of proprietary content. The outcome could impact the future of generative AI tools and their ability to use copyrighted material.
AI use rising in influence campaigns online, but impact limited - US cyber firm: Mandiant, a U.S. cybersecurity firm owned by Google, reports an uptick in the use of AI for manipulative online information campaigns. While AI-generated content has been employed in politically motivated influence campaigns, the impact remains limited. Generative AI’s potential to enhance influence campaigns is acknowledged, notably by groups with fewer resources. Though AI’s role in digital intrusions from major actors like Russia and China is currently low, experts anticipate its future growth and impact.
Snapchat's My AI chatbot glitched so hard it started posting Stories: Snapchat’s My AI chatbot, part of the Snapchat Plus package, experienced technical malfunctions, posting unexpected content to users’ feeds and becoming unresponsive. Users speculated about the AI’s capabilities, with some jokingly suggesting the rise of the Singularity. TechCrunch reported that the issue was a temporary outage, and a spokesperson clarified that My AI does not have the Stories feature. Despite the glitch, the chatbot offers various AI-powered features including AR filter recommendations and AI functionality in group chats, photos, and more.
AI Business & Market Trends
LinkedIn Says ChatGPT-Related Job Postings Have Ballooned 21-Fold Since November: Job postings related to AI, particularly GPT, and ChatGPT, have surged on platforms like LinkedIn, Indeed, and ZipRecruiter. LinkedIn’s report reveals a 21-fold increase in English-language job posts mentioning GPT or ChatGPT since November 2022. Positions like chief AI officer, AI engineer, and AI data scientist are emerging. The number of U.S. companies with a head of AI has nearly tripled in five years. As AI adoption grows, roles such as ethicists and chief trust officers are predicted to rise. LinkedIn users are adding AI skills to their profiles, indicating the increasing AI opportunities.
As Hollywood Strikes, 96% Of Entertainment Companies Are Boosting Generative AI Spend: As the Hollywood writers’ strike continues, a new report reveals that almost every entertainment and media company, along with other industries, plans to increase their spending on generative AI. According to the report by Lucidworks, 96% of executives prioritizing AI decisions are investing in generative AI, with China leading at 100%. The rapid growth of generative AI presents both opportunities and challenges, potentially disrupting various industries. The emergence of AI-generated synthetic actors also adds complexity to the entertainment industry, with AI-generated “talent” posing new possibilities and concerns.
OpenAI acquires Global Illumination: OpenAI has acquired the team from Global Illumination, a company known for leveraging AI to develop creative tools and digital experiences. The team, which includes individuals with experience from Instagram, Facebook, YouTube, Google, and more, will be working on OpenAI’s core products, including ChatGPT.
AI in Entertainment & Content Generation
New Startup Sakana AI Wants To Build Nature-Inspired Artificial Intelligence: Tokyo-based startup Sakana AI, co-founded by former Google AI experts Llion Jones and David Ha, aims to revolutionize AI models by drawing inspiration from nature. Jones, a co-author of the transformative 2017 Google paper “Attention Is All You Need,” left Google to serve as CTO, while Ha, previously heading Google’s AI research in Japan, will be CEO. Sakana aims to create adaptable AI models based on observations of natural systems, fostering flexibility and coordination. The founders envision interconnected smaller models working collaboratively, departing from the centralized approach of massive transformer models.
Runway launches new 'Watch' feature as CEO says Hollywood AI discourse 'needs to be more nuanced' Runway, a generative AI video startup, faces Hollywood skepticism due to the impact of generative AI on the entertainment industry. CEO Cristobal Valenzuela, while acknowledging concerns, believes the discourse needs nuance. Runway’s new “Watch” feature, launched to support artists and filmmakers, reflects the company’s drive to democratize and promote AI-generated content. Valenzuela emphasizes that AI tools require human agency and experimentation, advocating for diverse perspectives to shape the evolving intersection of art and technology.
Google said to be testing new life coach AI for providing helpful advice to people: Google’s DeepMind is reportedly testing new AI technology that could turn generative AI into a “personal life coach,” offering advice on personal and professional challenges. While this concept contradicts Google’s current privacy guidelines, it may be aimed at competing with rivals like OpenAI in the AI arms race. DeepMind has assembled a team of experts to evaluate the tool’s responses, including intimate questions about people’s lives. Despite the potential, it’s unclear whether this AI life coach will become a product or remain an experimental project.
AI in Research & Science
Energy announces $16M for AI nuclear physics research: The U.S. Department of Energy is allocating $16 million for 15 projects focused on leveraging AI and machine learning for advancements in nuclear physics research. These initiatives encompass nuclear decay research, optimizing accelerator beams, and detector design for the Brookhaven National Laboratory’s Electron-Ion Collider project. The funding aims to harness AI’s potential in expediting experimental discoveries in nuclear physics and addressing technical challenges in simulations, data acquisition, and analysis.
Ex-Google CEO Eric Schmidt to launch AI-science moonshot: Former Google CEO Eric Schmidt is launching an ambitious AI-driven initiative called ‘Semafor’ to address scientific challenges. The non-profit project, similar to OpenAI, aims to attract top talent in science and AI. Schmidt has hired scientists Samuel Rodrigues and Andrew White to lead the effort, which aims to leverage AI for breakthroughs in various fields, from drug discovery to material sciences. Funding will come from Schmidt’s personal wealth, and the project aims to provide competitive salaries and resources.
🎧 Did you know AI Breakfast has a podcast read by a human? Join AI Breakfast team member Luke (an actual AI researcher!) as he breaks down the week’s AI news, tools, and research: Listen here
5 new AI-powered tools from around the web
Featured Tool:
Eightify extracts key ideas from long Youtube videos. It uses GPT and their own technology to improve the quality of summaries and to support even 32-hour videos. Plus, the average generation time is just 8 seconds.
Speechflow is an AI model converting audio to punctuated text, primed for readability. It has 14 languages supported, claims 20% higher accuracy than competitors, and can handle 1-hour audio in <3 mins at $0.0002/sec. Streamlined API, and has effortless integration with concise code snippets.
Hurd.ai leverages AI to enable attentive listening during lectures, workshops, and meetings. Sit back while Hurd.ai captures, transcribes, summarizes, and tags your content. Unlike others charging by the hour, its free plan offers unlimited, secure recordings.
Roll is a robust AI video platform, that enables users to craft captivating videos using only a smartphone and browser. It boasts cinema-grade visuals, top-notch audio, multi-camera and VFX effects, audio edits, a Roll Editor, and universal publishing. By harnessing AI and machine learning, Roll automates video production for rapid, cost-effective professional-level results.
VectorShift is a no-code platform that lets companies build generative AI applications. Access pre-built pipelines like document generators, chatbots, and more, or craft custom solutions.
arXiv is a free online library where scientists share their research papers before they are published. Here are the top AI papers for today.
The paper introduces a groundbreaking approach to generate highly detailed and realistic 3D-clothed human models from single images. The method leverages a combination of descriptive text prompts and a personalized Text-To-Image (T2I) diffusion model for accurate reconstruction. By extracting both visible and indescribable details, TeCH achieves intricacies, textures and geometry. It employs a hybrid 3D representation and Score Distillation Sampling (SDS) for optimization. TeCH surpasses existing methods in rendering quality and geometry accuracy, exhibiting potential for various applications. This innovative technique bridges the gap between textual guidance and 3D human generation, paving the way for broader applications beyond human bodies.
The paper presents a pioneering approach that identifies pivotal states in reinforcement learning (RL) from video data. Addressing challenges in understanding RL decision points, the Deep State Identifier framework includes a return predictor and a critical state detector. This innovative method extracts significant information from visual trajectories, improving agent behavior and policy comparison. Unlike prior work, it operates without explicit action information, enhancing interpretability in RL. Extensive experiments validate its efficacy in diverse environments, underscoring its potential for hierarchical RL. The approach offers a unique perspective on explainable RL, showcasing its potential for improving agent performance and simplifying explanation processes.
The paper introduces the GPT-4 Code Interpreter (GPT4-Code) for solving complex math word problems. GPT-4 Code generates and executes Python code to enhance reasoning and problem-solving. The study analyzes its code usage frequency and reveals its self-debugging abilities, attributing its success to dynamic code generation. A novel explicit code-based self-verification (CSV) technique is proposed, guiding GPT-4 Code to verify and rectify solutions using code. The approach achieves impressive zero-shot accuracy (53.9% → 84.3%)on the MATH dataset, outperforming previous methods. This research advances the understanding of code-enhanced reasoning in LLMs and contributes new datasets to improve mathematical capabilities.
Crafted by a team of researchers from Google and the University of Michigan, this approach introduces a versatile methodology to imbue LLMs with the ability to personalize content creation. Mirroring the stages of writing instruction, the novel framework encompasses a multi-stage architecture encompassing retrieval, ranking, summarization, synthesis, and generation. Moreover, the model’s reading proficiency is enhanced through a multitask setting, similar to the interconnectedness of reading and writing skills in education. Experimental results across diverse datasets affirm the approach’s efficacy, promising a new era of tailored text generation.
Researchers from Zhejiang University, MIT Media Lab, and Image Derivative Inc. introduce a pioneering approach to construct relightable and animatable neural avatars from sparse-view (or monocular) videos of dynamic humans. The novel framework enables the generation of avatars that are not only animatable but also adaptable to different lighting conditions, a challenging feat. Through a clever Hierarchical Distance Query (HDQ) algorithm, the methods efficiently compute surface intersections and light visibility, facilitating realistic relighting. Experiments showcase the approach’s superior performance, offering a promising avenue for dynamic and realistic virtual human representation.
Thank you for reading today’s edition.
Your feedback is valuable.
Respond to this email and tell us how you think we could add more value to this newsletter.