Brief

Rox, a new sales platform, has launched utilizing OpenAI's API to enhance sales team productivity. The platform employs a swarm of AI agents powered by various OpenAI models, including GPT-4o and GPT-4o mini, to automate tasks, unify customer data, and provide actionable insights. Early adopters have reported significant gains, such as an average of 8 hours saved weekly per representative and a 35% increase in customer engagement, leading to a doubling of the sales pipeline. AI

Lindy AI, a startup in the AI agents space, has shifted its platform design from a single large text field to a visual workflow builder. This change aims to improve reliability and user experience by "putting agents on rails" rather than relying solely on raw LLM capabilities. The company's founder noted that current models like Claude 3.5 Sonnet are no longer the primary bottleneck, with integration quality and user workflow design becoming the new challenges. Lindy AI's founder also proposed that horizontal agent platforms, similar to search engines, may eventually dominate over specialized vertical platforms due to shared core functionalities. AI

Dust.tt, an AI infrastructure platform, has evolved from a developer framework and browser extension into a tool for deploying AI agents within enterprises. The company emphasizes a horizontal approach, enabling users to build their own AI experiences rather than focusing on specific vertical workflows. This strategy has led to high user engagement, with some deployments seeing 88% daily active users, and allows for emergent use cases driven by business needs. However, this horizontal strategy also presents challenges in go-to-market efforts and requires complex infrastructure to manage diverse integrations. AI

GitHub Copilot has introduced a new feature that allows users to create custom AI assistants tailored to specific projects or coding styles. These assistants can be trained on a user's codebase, enabling them to provide more context-aware and personalized code suggestions. This move aims to enhance developer productivity by offering more specialized AI support directly within the coding environment. AI

Decagon, a customer support automation company, is leveraging a suite of OpenAI's models, including GPT-3.5, GPT-4, GPT-4o, and GPT-4 Turbo, to provide fully automated customer service. The platform handles millions of support conversations for businesses globally, with one client seeing 91% of its global support managed without human intervention. Decagon's approach involves fine-tuning specific models for tasks like query rewriting and using others for complex decision-making, enabling rapid deployment and customization for clients. AI

Google's NotebookLM has introduced an innovative "Audio Overviews" feature that transforms documents and videos into conversational podcasts. Unlike typical text-to-speech summaries, this feature incorporates natural human interjections and pauses, creating a more engaging listening experience. The development team prioritized product intuition and user feedback, emphasizing simplicity and iterative improvement over complex customization options. This approach highlights the success of integrating AI engineering with product management to create unique and user-centric AI applications. AI

OpenAI announced its new Realtime API, enabling developers to build applications with near-instantaneous responses. This advancement is expected to significantly improve user experiences in areas like gaming and interactive AI assistants. The company also unveiled other developer-focused tools and updates during its recent Dev Day event. AI

OpenAI has launched a public beta of its Realtime API, allowing developers to integrate low-latency, speech-to-speech conversational experiences into their applications. This new API, powered by the GPT-4o model, enables natural interactions by streaming audio inputs and outputs directly, supporting features like interruption handling and function calling. Additionally, OpenAI is introducing audio input and output capabilities to its Chat Completions API, offering a simpler way for developers to build voice-enabled applications without needing to stitch together multiple models. AI

OpenAI has launched a new Model Distillation feature within its API, simplifying the process for developers. This new offering allows users to fine-tune more cost-efficient models, such as GPT-4o mini, using the outputs from larger, frontier models like GPT-4o. The integrated workflow includes Stored Completions for dataset generation and Evals for performance measurement, streamlining what was previously a complex, multi-step process. AI

OpenAI has launched an advanced voice mode for ChatGPT, enabling more natural and responsive conversations. This feature allows users to interact with ChatGPT using their voice, receiving spoken responses in real-time. The update aims to enhance user experience by making interactions more fluid and intuitive, similar to human conversation. AI

Genmab, a biotechnology firm, has launched an initiative called "AI Everywhere" to integrate OpenAI's ChatGPT Enterprise across its workforce of over 2,000 employees. This program aims to enhance efficiency and innovation by providing employees with AI tools for tasks ranging from drafting documents to complex data analysis. Employees report saving an average of 3.5 hours per week, and the company has developed over 100 custom GPTs, including tools for translating scientific documents and drafting clinical trial reports, leveraging advanced features like GPT-4o's vision capabilities. AI

Hugging Face has launched new Community Tools within its HuggingChat platform, allowing users to integrate custom tools and fine-tune models for specific tasks. This feature aims to enhance the chatbot's capabilities by enabling developers to create and share specialized functionalities. The integration of these tools is expected to broaden the application of large language models in various domains. AI

The AIPhone 16 is presented as a "Visual Intelligence Phone," suggesting a focus on integrating AI capabilities, particularly those related to visual processing, into a smartphone. This concept implies enhanced features for image recognition, augmented reality, or other visually-driven AI applications. The announcement positions the device as a significant step forward in mobile AI. AI

Replit has launched its AI agent, a tool designed to assist developers with coding tasks. This release comes as a surprise, as it appears to have beaten the highly anticipated Devin AI agent to market. The Replit Agent aims to streamline the development process by offering intelligent coding assistance. AI

Ada, an AI-native customer service automation platform, has rebuilt its product using OpenAI's GPT-4 and GPT-4o models to improve customer query resolution. The company found that while high containment rates were easy to achieve, customer satisfaction was often poor. Ada developed a new evaluation framework that measures how well conversations are resolved, aiming for a new industry standard beyond simple containment. AI

Upwork has announced a strategic partnership with OpenAI, integrating the company's AI models across its platform to enhance productivity and product offerings. This collaboration has led to the development of several new features, including a Job Post Generator that reduces creation time by 80% and Upwork Chat Pro powered by GPT-4o for freelancers. The company has also launched Uma, an AI companion that combines Upwork's data with OpenAI's models to assist clients and freelancers, aiming to improve user experience and increase engagement on the platform. AI

The AI news roundup highlights recent developments in large language models, specifically mentioning Grok 2 and an updated version of ChatGPT-4o. The article suggests that the latest iteration of ChatGPT-4o has caused some confusion among users or observers. Further details on the capabilities or implications of these models are not provided in the summary. AI

Smol AINews has released its latest newsletter, "Gemini Live," focusing on updates and news related to Google's Gemini AI models. The newsletter provides a curated selection of information for those interested in the advancements and applications of Gemini. It aims to keep readers informed about the rapidly evolving landscape of AI, specifically concerning Google's flagship AI technology. AI

Hugging Face has introduced a unified API for tool use in language models, aiming to simplify how developers integrate external tools and functions into their AI applications. This new approach standardizes the process, allowing models to more effectively interact with APIs and perform complex tasks. The update is expected to enhance the capabilities of AI assistants and chatbots by enabling seamless access to real-world data and services. AI

Rakuten, a global e-commerce and fintech giant, is partnering with OpenAI to leverage generative AI for enhanced customer insights and operational efficiency. The company is integrating OpenAI's APIs, including GPT-3.5 and RAG, to process vast amounts of transactional and unstructured data. This integration has already led to faster customer service responses and improved methods for summarizing product reviews, with plans to further utilize AI for B2B insights and new conversational experiences. AI

Smol AINews has reported on upcoming enhancements to OpenAI's GPT-4o model, slated for release in August. These updates are expected to introduce 100% structured output capabilities for all users, including a "mini" edition of the model. This feature aims to provide more predictable and usable outputs for developers and applications. AI

OpenAI has introduced Structured Outputs for its API, enabling models to reliably adhere to developer-provided JSON Schemas. This new feature enhances the accuracy of AI-generated structured data, a common use case for applications like data extraction and agentic workflows. The system achieves 100% reliability in evaluations with its latest GPT-4o model, significantly outperforming previous versions. Structured Outputs can be integrated via function calling with "strict: true" or through a new "json_schema" option in the "response_format" parameter for GPT-4o models. AI

OpenAI has launched a prototype called SearchGPT, which integrates AI models with real-time web information to provide direct and timely answers. This new search feature aims to improve the user experience by offering conversational follow-ups and clear sourcing from publishers. The company is collaborating with content creators to ensure their work is valued and discoverable within the AI-powered search results, with plans to incorporate successful features into ChatGPT. AI

Hugging Face has introduced TGI Multi-LoRA, a new feature for its Text Generation Inference (TGI) solution. This enhancement allows users to serve up to 30 different LoRA (Low-Rank Adaptation) models simultaneously from a single deployment. This significantly improves efficiency and reduces the computational resources needed for serving multiple specialized models. AI

Prezi is utilizing Hugging Face's platform and expert support to advance its machine learning initiatives. This collaboration focuses on integrating multimodal capabilities into Prezi's products, aiming to enhance user experience and streamline content creation. By leveraging Hugging Face's resources, Prezi seeks to accelerate its development cycle and stay at the forefront of AI-driven innovation. AI

Google's Gemini has reportedly introduced context caching, a feature designed to improve the efficiency of large language models by storing and reusing previously processed information. However, there is some uncertainty regarding the exact implementation and effectiveness of this new capability. The development aims to enhance Gemini's performance in handling long conversations or complex tasks by reducing redundant computations. AI

Paf, an international gaming company, has significantly boosted developer productivity by implementing custom GPTs built on ChatGPT Enterprise. Their engineering team created over 85 specialized GPTs to automate coding tasks, such as generating boilerplate code and converting API definitions, which has reduced errors and accelerated development cycles. Additionally, Paf is integrating this AI solution into its grit:lab coding academy to train new developers with an AI-augmented, systems-architecture mindset from the outset. AI

Clay, a sales intelligence platform, has experienced significant growth by integrating OpenAI's GPT-4 to create an AI agent named Claygent. This agent automates the process of researching and enriching sales leads by scraping and summarizing information from websites, a task previously handled manually by sales teams. Claygent's efficiency and accuracy, achieved through optimized token usage and cross-verification, have enabled a single person to perform the work of an entire team, contributing to Clay's 10x year-over-year revenue growth. AI

Color Health is integrating OpenAI's GPT-4o model into a new copilot application designed to accelerate cancer treatment for patients. This tool analyzes patient data, identifies missing diagnostics, and generates personalized screening and workup plans for clinicians to review. The application aims to streamline the complex and time-consuming process of cancer care, potentially reducing mortality risks associated with treatment delays. AI

Hugging Face Accelerate has introduced new integrations with DeepSpeed and Fully Sharded Data Parallel (FSDP). This update allows users to seamlessly switch between these two popular distributed training frameworks. The goal is to provide greater flexibility and performance optimization for large-scale model training. AI

Hugging Face has launched NPC-Playground, a new 3D environment designed for interacting with large language model-powered non-player characters. This tool allows users to create and engage with AI-driven characters in a virtual space, facilitating more dynamic and interactive experiences. The platform aims to bridge the gap between AI language models and interactive 3D environments, opening up new possibilities for game development and virtual world creation. AI

Anthropic has released a new feature called Tool Use GA, which allows its AI models to interact with external tools and APIs. This capability enables the models to perform a wider range of tasks, such as accessing real-time information or executing specific functions. The release aims to enhance the utility and versatility of Anthropic's AI offerings for developers and users. AI

MavenAGI has launched an AI-powered customer support agent utilizing OpenAI's GPT-4 model. This new service aims to address the common frustrations experienced by both customers and support agents by providing faster, more personalized, and higher quality assistance. The platform ingests company data, integrates with CRMs, and uses GPT-4's reasoning capabilities to answer questions and perform actions, with a self-evaluation mechanism to ensure accuracy. AI

Cursor has achieved a finetuning speed exceeding 1000 tokens per second for the Llama3-70b model. This advancement significantly accelerates the process of adapting large language models for specific tasks, such as fast file editing within the Cursor IDE. The improved finetuning capability aims to enhance developer productivity by making AI-powered code assistance more responsive and efficient. AI

Canva has reported that its AI-powered Magic Studio has been used over 5 billion times, with its Magic Write feature alone generating more than 10 billion words. The visual communication platform integrated OpenAI's GPT-4 API to enhance tools like Magic Write for text generation and summarization, Magic Design for prompt-based content creation, and Magic Switch for format conversion and translation. This collaboration aims to accelerate the creative process by making AI tools accessible within Canva's user-friendly interface, while also adhering to OpenAI's safety guidelines. AI

Hugging Face has released Transformers Agents 2.0, an updated framework designed to enable the creation of AI agents that can interact with tools and execute tasks. This new version aims to simplify the development process for building sophisticated AI agents, allowing them to leverage large language models for planning and decision-making. The update focuses on enhancing the agent's ability to utilize external resources and perform complex operations. AI

OpenAI and Stack Overflow have announced a strategic partnership focused on integrating Stack Overflow's vast knowledge base into OpenAI's AI models. This collaboration will allow OpenAI users, including those using ChatGPT, to access vetted technical information and code directly from Stack Overflow, with proper attribution. In return, Stack Overflow will leverage OpenAI's models to enhance its own OverflowAI product, aiming to improve developer experience and foster community engagement through AI. AI

Moderna has partnered with OpenAI to integrate ChatGPT Enterprise across its operations, aiming to accelerate the development of mRNA medicines. The pharmaceutical company has focused on a comprehensive workforce transformation program to ensure widespread adoption and proficiency of generative AI tools. This initiative builds upon their earlier success with an internal AI chatbot, mChat, which was developed using OpenAI's API and achieved over 80% employee adoption. AI

OpenAI has introduced several new enterprise-focused features for its API customers, aiming to enhance security, control, and cost management. New additions include Private Link for secure Azure-OpenAI communication, native Multi-Factor Authentication, and a Projects feature for granular oversight of API keys and model access. The Assistants API has also been updated with improved file ingestion limits, streaming support, and better cost controls, alongside new options for discounted usage and asynchronous workloads via a Batch API. AI

The Instructor library, an open-source SDK, has gained significant traction by simplifying the process of extracting structured data from large language models. It integrates with various LLM provider SDKs, allowing developers to define expected output structures using Pydantic models. Instructor supports key use cases such as extracting structured data from unstructured text, identifying complex relationships to form graphs, and improving query understanding for systems like RAG. This approach facilitates more controllable agent workflows and easier model swapping, moving beyond simple prompt engineering techniques. AI

Udio, a new product for AI-generated music, is highlighted as a key development in the burgeoning field of multi-modal AI. This technology represents a significant step forward in combining different data types to create complex outputs. The discussion also contrasts these advanced multi-modal approaches with older methods of integrating diverse data. AI

Sonauto has released a preview of its v3 AI music creation tool, which can generate full-length songs up to 4.5 minutes long. The tool aims to turn user ideas into songs rapidly, offering thousands of new styles. While in preview, v3 may occasionally produce lower-quality results. AI

Hugging Face has released an update to its Optimum Intel library, enhancing the performance of SetFit models on Intel Xeon processors. This optimization significantly speeds up inference times, making it more efficient to deploy these models in production environments. The improvements leverage specific hardware features of Xeon CPUs to achieve these gains. AI

Replit has developed a new AI model specifically trained to understand and operate within its development environment, aiming to enhance developer tools. This model's initial application is code repair, leveraging the vast amount of data from Language Server Protocol (LSP) diagnostics generated daily on the platform. The system reconstructs project states using Operational Transformations and synthesizes diffs with large language models to generate and verify code fixes. AI

Oscar Health has partnered with OpenAI to integrate AI into its health insurance operations, aiming to reduce costs and enhance patient care. The company is utilizing OpenAI's models, which it found to perform best on healthcare-specific tasks, to automate clinical documentation and claims processing. These AI applications have significantly reduced the time spent on these tasks, with potential for further productivity gains, and are also being used to analyze complex medical records for better patient management and equity. AI

Zelma, a new research assistant powered by OpenAI's GPT-4, aims to make U.S. education data more accessible to parents, teachers, and policymakers. Developed by economist Dr. Emily Oster and her team at Brown University, Zelma processes and visualizes standardized test performance data, which is often scattered and difficult to access. The tool utilizes function calling and fine-tuning to allow users to ask questions in plain language and receive tailored insights and data visualizations, with explanations of the underlying logic and context. AI

Researchers have explored interventions on a language model trained to play chess, dubbed Chess-GPT. By manipulating the model's internal representations of the board state and player skill, they demonstrated a causal link between these representations and the model's output. This work addresses skepticism about whether large language models possess genuine world models or merely learn superficial patterns, showing that targeted edits can influence the model's playing strength and move generation. AI

Hugging Face has launched the Chatbot Guardrails Arena, a new platform designed to evaluate the safety and ethical alignment of large language models. This arena allows developers and researchers to test how well chatbots adhere to safety guidelines and identify potential risks. The initiative aims to foster more responsible AI development by providing a transparent and collaborative environment for safety testing. AI

Superhuman has integrated OpenAI's API to enhance its email client, introducing AI-powered features designed to significantly reduce the time professionals spend managing their inboxes. These new capabilities include AI-assisted email composition, voice-to-email generation, automatic summarization, and one-click replies, with over 85% of users adopting the AI features. The company reports that these tools are doubling inbox processing speed and email writing speed for users. Superhuman's CTO also discussed the potential for inboxes to become central AI agents, leveraging vast amounts of personal data for proactive assistance. AI

Hugging Face has released new resources and guides detailing how to leverage Intel's Gaudi 2 AI accelerators for efficient AI model training and deployment. These collaborations focus on optimizing performance for tasks like assisted generation and Retrieval-Augmented Generation (RAG) applications, aiming to provide cost-effective solutions for enterprises. The initiative also explores running generative AI models on Intel's CPU and Xeon processors, broadening the accessibility of AI hardware. AI