AI News & Launches

Accorata is an AI-powered platform providing early-stage deal sourcing, helping investors track global pre-seed and seed startups rapidly. It integrates directly with inboxes to filter relevant decks and uses AI for founder due diligence, assuring swift access to investment opportunities from over 30 sources online.

ElevenLabs introduces Texts to Sounds Effects API, which is now available for users to integrate into their applications. The API converts text inputs into sound effects, enhancing the interactive and multimedia experience. The implementation code is open-sourced and available on GitHub.

Analysis predicts that if the growth of language model dataset sizes continues, the total stock of high-quality human-generated text data (~300T tokens) could be fully utilized between 2026 and 2032.

Cohere introduces Command R+, their newest and most powerful model in the Command R family, designed to automate complex business workflows. The multi-step tool leverages the use of AI to simplify and enhance productivity.

Redwood Research outlines achieving 50% accuracy on ARC-AGI tests using GPT-4o. The team generated about 8,000 Python programs per problem, enhancing the AI's performance using few-shot prompts, program revisions, and better grid representations compared to previous methods.

Spain's National Institute of Social Security (INSS) utilizes secretive AI systems to detect potential fraud in sick leave benefits, raising questions about transparency, data privacy, and algorithmic discrimination. An in-depth investigation uncovers the limited success of these algorithms and their impact on both the public and medical inspectors.

TokenCost is an MIT-licensed project providing easy token price estimates for over 400 Large Language Models (LLMs). It helps calculate USD cost for prompts and completions across major LLM APIs, including token counting and price tracking for updated model prices.

The essay discusses the concept of intelligence, proposing it's a subjective, human-centered label rather than an empirical quality. Intelligence is seen as a collection of traits that have helped humans survive but is not necessarily central to the success of other life forms on Earth. The authors argue that intelligence should be recognized as highly relative, influenced by individual perspectives, and that other forms of life should be appreciated for their own unique strategies of success.

Amazon's Rekognition AI was used in UK train stations to analyze passengers’ age, gender, and emotions, sparking privacy issues. The undisclosed trials raise concerns about AI's normalization in public spaces without public discourse, with technology experts questioning the reliability of emotion detection and warning against possible abuses.

The article by Jeremy Howard discusses the nuances of AI policy, specifically the differences between 'deployment' and 'release' of AI models, with a focus on California's SB 1047 bill. It argues for a focus on regulating the deployment of AI systems rather than on restricting the release of models to ensure safety without stifling open-source innovation.

Warp is an AI-powered terminal application providing an enhanced command-line experience with features like Agent Mode, which understands natural language for tasks and fixes. It's available for Mac and Linux, with a Windows version coming soon.

Florida Sheriff Grady Judd announced the launch of the Sheriff's Artificial Intelligence Laboratory (SAIL) in partnership with Florida Polytechnic University to utilize AI in crime fighting, a move fraught with criticism given Judd's questionable law enforcement practices.

Proton, known for its secure email and productivity suite, is transitioning into a nonprofit foundation, distancing itself from models reliant on big donors, Google, or government funding. Marking 10 years from its Proton Mail crowdfunding, CEO Andy Yen pledges focus on privacy over profit, bolstered by Switzerland's shareholder-free foundation laws, ensuring adherence to their founding mission.

Apple CEO Tim Cook expresses uncertainty about completely preventing AI hallucinations in new Apple Intelligence system, while remaining confident in its quality. Apple's WWDC reveal, OpenAI partnership, and possible future collaborations are also discussed.

Drawbert AI is an innovative website that utilizes artificial intelligence to transform simple photos into stunning portrait artworks. It offers the ability to convert individual photos into paintings or create artistic group portraits efficiently.

The article discusses the nature of discussions on AI bias, focusing on how people perceive bias in machine learning (ML) differently compared to classical bugs in software. Drawing examples from various AI platforms and forums, it highlights conflicting views on whether certain AI behaviors represent bias or whether they are a result of data representation and model limitations.

McDonald's is ceasing its AI-driven drive-thru test, a collaboration with IBM, halting the technology in over 100 sites by July 26, 2024. Despite this, the company aims to continue exploring voice ordering solutions with a potential new partner.

Epipheo AI is an AI-powered tool that simplifies the video creation process, enabling users to generate professional explainer videos in minutes. It offers customizable voiceovers, various visual styles, and quick iteration for editing and sharing. No credit card is required to get started.

Meta halts AI training using data from EU Facebook and Instagram posts, responding to European regulators' privacy concerns, despite labeling it a setback for innovation.

Wunjo is an application for neural synthesis with support for both Windows and Ubuntu platforms. It provides professional and community editions, requiring significant system resources, such as VRAM and RAM. The software features automatic dependency and models installation, with detailed tutorials available for setup. Upon first launch, it downloads necessary models and offers a GPU version that leverages Nvidia cards with CUDA 11.8 for improved performance.

Olvy is a powerful platform that leverages AI technology to help businesses manage and analyze customer feedback efficiently. It offers a unified repository for feedback from multiple channels, allows for in-depth analysis with thematic and sentiment tools, and helps prioritize actions with smart feedback categorization. Companies like LottieFiles and Ready Player Me have significantly increased productivity and operational efficiency using Olvy.

Adam Unikowsky's Substack post, 'In AI we trust, part II', discusses the use of AI to adjudicate Supreme Court cases, citing empirical tests with Claude 3 Opus (comparable to GPT-4) on this term's cases, revealing AI's high accuracy and efficiency in legal analysis, potentially outperforming human law clerks.

McDonald’s is ending its AI-driven drive-thru ordering test program with IBM by July 26th, 2024. The project had examined if an AI voice ordering chatbot could speed up drive-thru services. With uncertain reasons for the IBM partnership halt, McDonald's still sees a future for voice ordering solutions in its restaurants.

Meta has achieved milestones in scaling AI capacities, including transforming its GPU clusters to adapt to the computational demands of large generative AI models. With plans to scale to 600,000 GPUs, Meta meticulously maintains AI training clusters through specialized hardware, up-to-date software, and multiple maintenance operations without disrupting availability.

The Economist has launched its statistical forecast for the 2024 US presidential election, applying a model developed with political scientists, including Andrew Gelman. The method combines national polls, economic indicators ('fundamentals'), and machine learning techniques to predict probabilities of winning for candidates in each state. Using Bayesian inference, the model updates its predictions with new polling data, while accounting for polling errors and biases. This comprehensive model will provide daily updates until election day.

NVIDIA has unveiled the Nemotron-4 340B model, an advanced synthetic data generator for training large language models, which showcases impressive capabilities and commercially viable licensing, setting a benchmark in the AI industry and rivalling the performance of OpenAI's GPT-4.

Drip is an AI-driven journaling app that empowers users to engage in deep introspection for authentic growth. It offers personalized prompts, secure and private reflection, and is designed with user-friendly features to enhance the journaling experience.

MechanicBotAI offers an AI-powered app to diagnose your car's issues. Users can describe their car's problems, enter details, and receive instant diagnoses to save money and time.

Inbox Zero is an AI-powered email management tool designed to help users achieve an organized inbox effortlessly. By prioritizing important emails and summarizing content, it streamlines your workflow and saves time.

Robb Knight details his experience with Perplexity AI ignoring his robots.txt rules and disguising its user agent. Knight started blocking 'PerplexityBot' via robots.txt on March 30th and server-side via nginx on June 14th. Despite these measures, Perplexity AI provided summaries of content that should have been restricted. Knight discovered Perplexity was scraping content using headless browsers without identifying as 'PerplexityBot', bypassing robots.txt. His further investigation showed Perplexity using a generic Chrome user agent string, indicating dishonest practices in content access.

Amazon has harnessed a grey-area strategy to gather data from Microsoft's GitHub for their AI development by encouraging employees to create multiple accounts for scraping purposes. This raises concerns over data privacy and the exploitation of platform resources for competitive advantage.

DESSIGN is a hub for designers, developers, and marketers featuring curated resources, templates, and tools. It offers a wide range of products and information to enhance digital projects and streamline production workflows.

Submit AI Tool is an online platform where users can discover a variety of artificial intelligence tools. The website features a wide range of AI tools tailored to different needs, including productivity, marketing, video, design, coding, and research. Users can also find the latest AI technology trends, new launches, and verified tools.

React Flow mini map feature allows users to navigate through nodes and edges with keyboard interactions, providing an intuitive way to manage workflows.

MindPal offers a 4-step workflow for conducting a thorough Competitive Analysis to understand market competitors and inform business strategy. The steps include identifying top competitors, analyzing them with SWOT, understanding market trends and regulations, and finale detailed analysis for growth opportunities.

inncivio is an innovative, AI-powered education platform designed to empower businesses by offering personalized content for skill enhancement. Aimed at facilitating learning for team members, inncivio addresses common business challenges such as engagement and uniqueness in training programs.

1000 Happy Dads is a blog that provides tips, inspiration, and resources for fathers to help them create memorable experiences and foster positive relationships with their children.

The post discusses optimizing a local LLM voice assistant using RAG (Retrieval Augmented Generation) to improve scalability and performance. Experiments showed reduced prompt sizes and faster response times without compromising the quality of interactions.

Chinese women find virtual romance with 'Dan' — a 'jailbreak' version of ChatGPT that bypasses OpenAI's safeguards, including NSFW content restrictions, gaining popularity among users seeking emotional support and companionship. With endorsements from influencers like Lisa Li, 'Dan' provides a personalized, available 24/7 partner experience, sparking ethical and privacy concerns.

OpenAI has appointed Paul M. Nakasone, a retired U.S. Army general and former director of the National Security Agency, as its newest board member. Nakasone brings a wealth of experience in cybersecurity, serving as the longest-tenured leader of the U.S. Cyber Command and chief of the Central Security Service.