Top News
OpenAI Introduces CriticGPT: A New Artificial Intelligence AI Model based on GPT-4 to Catch Errors in ChatGPT’s Code Output
OpenAI has introduced a new AI model, CriticGPT, designed to identify errors in the outputs of ChatGPT, an AI system built on the GPT-4 architecture. CriticGPT, which is trained via Reinforcement Learning with Human Feedback (RLHF), aims to address the challenges human reviewers face in consistently assessing the accuracy and quality of complex AI outputs. The model has shown significant effectiveness, with human reviewers using CriticGPT performing 60% better in evaluating ChatGPT’s code outputs than those without such assistance. As RLHF is essential to the performance of today’s chatbots, such improvements to the process could have a significant impact on ChatGPT’s quality.
Music labels sue AI music generators for copyright infringement
Universal Music Group, Sony Music, and Warner Records have filed lawsuits against AI music-synthesis companies Udio and Suno, accusing them of mass copyright infringement. The AI companies allegedly used copyrighted music to train their AI models, which can generate new songs based on text descriptions. The record labels argue that this could lead to AI-generated music that competes with and devalues the work of human artists. The labels are seeking damages of up to $150,000 per song used in training. This lawsuit could have significant implications for the future of generative AI in creative fields, potentially requiring companies to license all musical training data used in creating music-synthesis models.
Anthropic Debuts Collaboration Tools for Claude AI Assistant
Anthropic, the creator of the Claude AI assistant, has launched an update to enhance team collaboration and productivity. The update introduces a Projects feature, allowing users to organize their interactions with Claude, including chat activity and knowledge sets, into one place. The Projects feature also enables users to incorporate internal organizational knowledge into Claude’s responses by adding documents such as style guides, codebases, and interview transcripts. Additionally, the update includes custom instructions for each Project, a new feature called Artifacts for generating and viewing content, and the ability for Team users to share snapshots of conversations with Claude. Anthropic emphasizes that these new features aim to integrate Claude into existing team processes and assures that any data shared within Projects will not be used to train their generative models without user consent.
Waymo ditches the waitlist and opens up its robotaxis to everyone in San Francisco
Waymo, the autonomous vehicle company under Alphabet, has announced that its robotaxi service in San Francisco is now open to the public, eliminating the need for customers to sign up for a waitlist. The move mirrors the company’s operations in Phoenix, where the service has been available without a waitlist since 2020. This decision comes as Waymo seeks to solidify its position in the robotaxi industry, despite recent scrutiny following a series of crashes and complaints about obstructions and delays. The company’s expansion of its service to all San Francisco residents is seen as a crucial step towards the normalization of autonomous vehicles and a potential path to profitability for the historically money-losing operation.
Other News
Tools
Character.AI now allows users to talk with AI avatars over calls – Character.AI now allows users to talk to AI characters over calls in multiple languages, offering a seamless experience with reduced latency and the ability to switch between calling and texting.
Meta just dropped an open source GPT-4o style model – Meta has publicly released a new family of AI models called Chameleon, capable of understanding and generating images and text, and handling prompts that call for outputs with both text and images.
OpenAI’s ChatGPT for Mac is now available to all users – OpenAI’s ChatGPT app for macOS is now available to all users, offering a desktop window version of the web app with the ability to select between different GPT models and a system-wide keyboard shortcut for prompt input.
Google Translate Just Added 110 More Languages – Google Translate has added 110 new languages, including Cantonese and Punjabi, using its AI PaLM 2 large language model, bringing the total of supported languages to nearly 250.
Figma announces big redesign with AI – Figma announces a major UI redesign and new generative AI tools to help users easily create projects, along with built-in slideshow functionality and other AI features.
Opera’s browser adds AI-powered image generation and better multimedia controls – Opera’s browser introduces AI-powered image generation, improved multimedia controls, and split tabs in its second version, Opera One R2, along with new themes and design elements for a fresh look.
ElevenLabs Launches Reader, A Text-to-Audio App – ElevenLabs has launched a new Reader app that uses AI-generated voices to turn written content into audio, catering to situations where reading from a screen is impractical or unsafe.
WhatsApp to introduce feature allowing users to choose ‘Meta AI model’ – WhatsApp for Android is developing a new feature allowing users to choose the Meta AI (Llama) 3 model to power the Meta AI chatbot. According to WABetaInfo, this feature is still under development and not available to beta testers yet.
Video editing app Captions releases AI edit feature that automatically adds effects to your video – Video editing app Captions has launched a new AI edit feature that automatically adds custom graphics, zooms, music, sound effects, transitions, and motion backgrounds to unedited videos, aiming to simplify the video-making process and enable mass production of content.
Google is bringing Gemini access to teens using their school accounts – Google is introducing its AI technology Gemini to teen students using their school accounts, providing educators with new tools and emphasizing responsible use of the technology.
Synthesia announces platform update with interactive AI videos, full-body avatars – Don’t miss OpenAI, Chevron, Nvidia, Kaiser Permanente, and Capital One leaders only at VentureBeat Transform 2024. Gain essential insights about GenAI and expand your network at this exclusive three day event. Learn More Officially dubbed Synthesia 2.
Business
Agility Robotics’ Digit humanoids land first official job – Agility Robotics’ Digit humanoids have landed their first official job with GXO Logistics Inc., marking the industry’s first formal commercial deployment of humanoids and the first RaaS deployment of such robots, generating real revenue and solving real-world business problems.
Formation Bio raises $372M to boost drug development with AI – Formation Bio raises $372M in Series D funding to apply AI to drug development, aiming to streamline clinical trials and drug development processes, with a focus on partnership acquisition and R&D.
OpenAI buys a remote collaboration platform – OpenAI acquires Multi, a video-first collaboration platform, as part of its strategy to invest heavily in enterprise solutions, following its recent acquisition of database tech firm Rockset.
Time Magazine Partners with OpenAI and ElevenLabs to Embrace AI Technology – Time Magazine partners with OpenAI and ElevenLabs to leverage generative AI for content creation and audio narration, showcasing its commitment to embracing new technology and innovation in the media industry.
Meta starts testing user-created AI chatbots on Instagram – Meta is testing user-created AI chatbots on Instagram, allowing creators to engage with fans and customers through AI avatars.
NBC to use AI-generated version of Al Michaels’ voice during Summer Olympics – NBC plans to use an AI-generated version of Al Michaels’ voice to narrate daily streaming recaps of the Summer Olympics in Paris, stunning Michaels with its close resemblance to his own style.
AI startup Emergence AI raises a ton of cash to enhance office worker productivity – Emergence AI, a generative artificial intelligence startup that’s focused on enhancing the productivity of business employees, said today it has closed on a $97.2 million funding round led by Learn Capital.
Amazon hires founders from well-funded enterprise AI startup Adept to boost tech giant’s ‘AGI’ team – Amazon hires executives from Adept, a San Francisco-based startup, to boost its AI efforts, with Adept continuing to operate independently and Amazon using some of Adept’s technology under a non-exclusive license.
Andrew Ng plans to raise $120M for next AI Fund – Andrew Ng plans to raise $120M for his second AI Fund, which aims to back small teams of experts solving key problems using AI, despite a decline in generative AI dealmaking and enterprise reluctance.
OpenAI delays ChatGPT’s new Voice Mode – OpenAI delays the launch of ChatGPT’s advanced Voice Mode due to lingering issues and the need for improvements, with the feature not expected to roll out to all customers until the fall.
Stability AI lands a lifeline from Sean Parker, Greycroft – Sean Parker and other investors have injected fresh capital into beleaguered generative AI startup Stability AI, with Parker joining as executive board chairman, as the company faces financial ruin, unpaid cloud bills, and copyright infringement suits, and plans to focus on growing its managed image, video, and audio pipelines and workflows.
The A.I. Boom Has an Unlikely Early Winner: Wonky Consultants – Consulting firms like Boston Consulting Group are experiencing a surge in revenue and demand for their services as companies seek guidance on how to leverage artificial intelligence for their businesses.
Anthropic tries ‘to enable beneficial uses’ of AI by government agencies – Anthropic aims to enable beneficial uses of AI by government agencies, positioning itself as an ethical choice among rivals and offering its AI models in the AWS Marketplace for the US Intelligence Community and in AWS GovCloud.
Time strikes licensing deal with OpenAI – Time has struck a multiyear content licensing deal and strategic partnership with OpenAI. The deal gives OpenAI access to Time’s archives from the last 101 years to train its large language models and use for responses to user queries in its consumer-facing products, such as ChatGPT.
Research
Cambrian-1: A Fully Open, Vision-Centric Exploration of Multimodal LLMs – Introducing Cambrian-1, a vision-centric exploration of multimodal language models, addressing the need for stronger visual representation learning and offering new insights and benchmarks for improving multimodal systems.
Solving the ‘Lost-in-the-Middle’ Problem in Large Language Models: A Breakthrough in Attention Calibration – Researchers have developed a breakthrough attention calibration mechanism to address the “lost-in-the-middle” problem in large language models, significantly improving their ability to utilize mid-sequence information effectively.
Building on Efficient Foundations: Effectively Training LLMs with Structured Feedforward Layers – Training LLMs with structured feedforward layers is essential for effectively utilizing AI in various applications.
RouteLLM: Learning to Route LLMs with Preference Data – RouteLLM is a new AI model that learns to route Large Language Models (LLMs) using preference data.
MultiDiff: Consistent Novel View Synthesis from a Single Image – AI model MultiDiff is capable of synthesizing consistent novel views from a single image, demonstrating its potential for various applications.
Efficacy of Language Model Self-Play in Non-Zero-Sum Games – Language model self-play in non-zero-sum games has been embraced for its efficacy by individuals and organizations working with arXivLabs.
CodeGemma: Open Code Models Based on Gemma – CodeGemma introduces open code models based on Gemma, including specialized variants for code and natural language generation tasks.
CLIPAway: Harmonizing Focused Embeddings for Removing Objects via Diffusion Models – A new approach called CLIPAway uses CLIP embeddings to focus on background regions and exclude foreground elements, enhancing inpainting accuracy and quality for seamless object removal.
Google Releases Gemma 2 Series Models: Advanced LLM Models in 9B and 27B Sizes Trained on 13T Tokens – Google’s Gemma 2 27B and 9B models represent a significant advancement in AI language processing, offering high performance and efficiency for a wide range of applications.
Concerns
The Center for Investigative Reporting is suing OpenAI and Microsoft – Nonprofit sues OpenAI and Microsoft for alleged copyright infringement, claiming the companies used their stories without permission or compensation.
Amazon Is Investigating Perplexity Over Claims of Scraping Abuse – Amazon’s cloud division is investigating Perplexity AI for potentially violating Amazon Web Services rules by scraping websites that attempted to prevent it from doing so.
Perplexity Plagiarized Our Story About How Perplexity Is a Bullshit Machine – AI-powered search startup Perplexity is accused of plagiarism by Forbes, with evidence showing the company’s chatbot generating text closely summarizing a WIRED article it was unable to access, raising concerns of plagiarism.
Morgan Freeman Slams AI Voice Imitations of Himself, Thanks Fans for Calling Out the ‘Scam’ – Morgan Freeman expresses gratitude to fans for calling out unauthorized AI imitations of his voice, highlighting the growing issue of AI-generated voice imitations in the entertainment industry.
Policy
Beijing supports autonomous vehicles with biggest regulation since 2019 – Beijing is supporting the development of autonomous vehicles with a comprehensive regulation that includes allowing certain entities to conduct mapping services using self-driving cars and creating a regulated sandbox for testing new technologies.
Y Combinator rallies start-ups against California’s AI safety bill – Y Combinator and AI start-ups oppose California’s AI safety bill, arguing that it could hinder innovation and open-source AI development.
Analysis
AI scaling myths – AI scaling myths: The belief that AI scaling will lead to artificial general intelligence is based on misconceptions about scaling laws, the availability of training data, and the limitations of synthetic data, leading to skepticism about the potential for continued improvement through scaling.
Expert Opinions
Large Language Models (Likely) Aren’t the Future of Artificial General Intelligence, And That’s Okay – AGI aims to replicate human intelligence, including creativity, intuition, and ethical decision-making, which current Large Language Models (LLMs) like GPT-4 fall short of, but they can still be valuable as part of a modular and adaptable AI system.
Microsoft’s Mustafa Suleyman says he loves Sam Altman, believes he’s sincere about AI safety – Microsoft’s Mustafa Suleyman admires OpenAI CEO Sam Altman, trusts in their partnership, and advocates for cooperation with China and the use of AI in classrooms, while also emphasizing the need for regulation and a slower pace in AI development.
MIT robotics pioneer Rodney Brooks thinks people are vastly overestimating generative AI – MIT robotics pioneer Rodney Brooks believes that people are overestimating the capabilities of generative AI and that it’s flawed to assign human capabilities to it.
Source: Read MoreÂ
