In partnership with

OpenAI introduces new enterprise AI features

In response to Meta's latest advancements with Llama 3, OpenAI has launched new features for its enterprise customers. OpenAI has upgraded its Assistants API, which now includes enhanced file retrieval capabilities, able to handle up to 10,000 files per assistant—a substantial increase from the previous 20-file limit. This upgrade also brings improvements such as parallel queries, improved reranking, and query rewriting, along with the introduction of new security features like Private Link for secure Azure cloud communications, and native Multi-Factor Authentication (MFA) to meet compliance demands.

Further enhancing the enterprise experience, OpenAI introduced the 'Projects' feature, allowing more precise administrative control and cost management. This feature enables enterprises to isolate specific tasks or documents to certain teams within the organization, thereby enhancing security and operational efficiency. Additionally, OpenAI has introduced cost management features, including discounted rates for consistent token usage and a 50% reduction for asynchronous workloads through the new Batch API, which promises faster results, potentially within 10-20 minutes.

OpenAI's strategy focuses on providing robust, scalable solutions that simplify integration for businesses, countering the rising competition from Meta's Llama 3 and other open models like those from Mistral. The emphasis on security, cost efficiency, and administrative control illustrates OpenAI's commitment to maintaining its lead in the generative AI industry, even as it welcomes competition to foster innovation. These enhancements offer a more streamlined, secure, and cost-effective AI deployment, making OpenAI's offerings more attractive to enterprises looking to integrate advanced AI capabilities into their operations.

Two Forbes 30U30 Founders Transforming Mental Wellness

People scroll on their smartphones, they get depressed. Today, 25% of American adults are suffering from a diagnosable mental health condition.

After seeing their mother struggle with depression, Lee brothers, Forbes 30U30 winners, founded Aura, already with over 8 million users and 100K+ paying subscribers.

Aura quickly received investments from Silicon Valley’s legendary VCs as well as executives from Spotify, Youtube, Facebook, and Apple. And they’re just getting started.

For [x] more days, you can invest in Aura alongside 2000+ investors and back the future of mental wellness.

Only a few days left to invest.

The Ray-Ban Meta Smart Glasses have multimodal AI now

The Ray-Ban Meta Smart Glasses, originally launched as an innovative content capture device and headphone set, have recently been upgraded to include multimodal AI. This new feature allows the glasses to process multiple types of information such as photos, audio, and text, enabling users to perform a variety of tasks like identifying plants or translating signs. Despite some limitations and occasional inaccuracies, such as confusing different car models, the AI enhances the practical utility of the glasses in everyday situations.

Users interact with the AI through simple voice commands, asking the glasses to identify objects or provide information, which the AI retrieves from the cloud. While the AI doesn't promise to perform every imaginable task, it has proved particularly useful for identifying vehicles and various plants, albeit with mixed accuracy. This integration turns the glasses into a more interactive tool, reducing the need to constantly pull out a smartphone for quick queries while on the go.

The physical design of the glasses, similar to regular eyewear and paired with open-ear headphones, makes them a discreet and convenient platform for wearable technology. While not every task suits the form factor — for instance, tasks requiring indoor use or detailed zoom are less feasible — the glasses represent a step towards more integrated wearable technology. They remain useful as high-quality headphones and a POV camera, ensuring their value extends beyond just the new AI features.

Microsoft launches Phi-3, its smallest AI model yet

Microsoft has introduced Phi-3 Mini, the smallest in its new series of AI models designed for efficiency and reduced computational demand. The Phi-3 Mini, with 3.8 billion parameters, is available on platforms like Azure, Hugging Face, and Ollama, and is poised to be followed by slightly larger models within the same series. This launch follows the release of Phi-2, which was acclaimed for matching the performance of larger models despite its smaller size.

The development of Phi-3 Mini was notably influenced by a unique training approach akin to how children learn from bedtime stories, utilizing simpler language and structures to cover complex topics. Microsoft used a generated set of 'children's books' to train the model, providing a foundation that improves upon the coding and reasoning capabilities seen in its predecessors, Phi-1 and Phi-2.

As the trend towards smaller, more efficient AI models continues, Microsoft's competitors like Google and Anthropic are also developing compact models aimed at specific tasks such as chatbots, coding assistance, and document summarization. These smaller models not only reduce costs but also offer tailored solutions that perform well on personal devices and are well-suited to handle custom applications with smaller internal datasets.

Other stuff

• Wendy's is using FreshAI to automate drive thru ordering

• Fine-tune Llama 3 with ORPO

• Tutorial: Build an LLM RAG Chatbot With LangChain

• Adobe presents VideoGigaGAN - A new video upscaling model that can upsample a video up to 8x with rich details.

• Geoffrey Hinton calls on regulators to ban Open Source AI models.

• Apple Reportedly Developing Its Own Custom Silicon for AI Servers

• Ex-Amazon exec claims she was asked to ignore copyright law in race to AI

• Anthropic Prompt library

Maximize Efficiency with Custom Prompts in SuperPower ChatGPT: A Comprehensive Guide

Hope you enjoyed today's newsletter

⚡️ Join over 200,000 people using the Superpower ChatGPT extension on Chrome and Firefox.