OpenAI's GPT Store and GPT for Teams

Author

AI Breakfast
January 12, 2024

Good morning. It’s Friday, January 12th.

Did you know: The iPod Shuffle was released 19 years ago today?

In today’s email:

  • AI Development and Advancements

  • AI in Business and Corporate Sector

  • AI and Legal/Regulatory Issues

  • 8 New AI Tools

  • Latest AI Research Papers

  • ChatGPT Creates Comics

You read. We listen. Let us know what you think by replying to this email.

Interested in reaching 47,998 smart readers like you? To become an AI Breakfast sponsor, apply here. 

Today’s trending AI news stories

AI Development and Advancements

> OpenAI has launched two features for ChatGPT: The GPT Store for ChatGPT for Teams. The GPT Store enhances ChatGPT’s utility in specific scenarios and allows for easy navigation through a plethora of custom GPTs. ChatGPT for Teams is designed for up to 150 users and offers an administrative interface for easier software distribution among employees, priced at $25-$30 per month per person. This plan provides access to the 32K model of GPT-4, with an unspecified higher message limit. Additionally, OpenAI assures that chats from team accounts won’t be used for training, contrasting with ChatGPT Plus users who must forego chat history for privacy. The initiative targets smaller teams and companies, intending to integrate tools like GPT-4, DALL-E 3, and Advanced Data Analysis into their workflow.

> QuEra, a quantum computing startup formed by Harvard and MIT researchers, is ambitiously planning to reveal a 10,000-physical-qubit quantum computer with 100 logical qubits by 2026. This advanced system aims to show a “practical quantum advantage,” outperforming traditional binary computers in complex calculations. QuEra’s roadmap includes intermediate milestones, developing systems with 256 and 3,000 physical qubits in 2024 and 2025, respectively. Their unique approach utilizes atoms trapped by laser tweezers as qubits, differing from IBM’s superconducting charge qubits. QuEra’s technology focuses on overcoming quantum computing’s major challenge of error-correction, aiming to maintain quantum information integrity for extended periods.

> OpenAI’s commercial iteration of ChatGPT has swiftly garnered substantial corporate interest, amassing 260 enterprise clients within a mere four-month span. The enterprise customers, encompassing a workforce exceeding 150,000 individuals, signify a broad-based corporate embrace of the AI tool. Launched in August, ChatGPT touts enhanced capabilities and stringent privacy measures, such as data encryption and a policy against utilizing customer data for technology advancement. This disclosure, made by OpenAI’s COO Brad Lightcap, underscores the program’s flexible pricing model tailored to diverse business requirements.

> Liquid AI, an MIT spinoff co-founded by robotics expert Daniela Rus, has debuted an innovative AI model, the liquid neural network. Bolstered by a robust $37.5 million in seed funding and attaining a valuation of $303 million, the company is poised for rapid growth. Modeled after the neural structure of roundworms, these networks are more compact and require significantly less computational power than traditional AI models. They are uniquely adaptable, adjusting to environmental changes over time. In practical applications, such as autonomous drone navigation through challenging terrains, Liquid AI’s networks have demonstrated superior performance. Based in Palo Alto and Boston, Liquid AI is set to expand its current team of 12.

> In a significant leap for 3D modeling, a collaborative effort from ETH Zurich and Google Zurich researchers has led to the creation of InseRF. This pioneering technique uses text directives and 2D outlines to integrate objects into three-dimensional scenes. By combining the latest in Neural Radiance Fields and generative AI, InseRF skillfully embeds objects with remarkable consistency, without relying on explicit 3D data. Tested on various real-world scenes, InseRF demonstrates efficacy in modifying scenes with 3D-coherent objects, marking a promising direction for enhancing 2D and 3D models.

> NVIDIA has introduced “Chat with RTX,” a pioneering application that enables users to tailor a chatbot to their local files, ranging from documents to video transcripts. Utilizing advanced technologies like Retrieval Augmented Generation and TensorRT-LLM, coupled with RTX acceleration, this chatbot provides prompt, context-sensitive responses and runs securely on Windows RTX PCs. This development represents a notable step forward in chatbot capabilities, offering users the flexibility to incorporate a variety of file formats and even transcripts from YouTube playlists, enhancing the chatbot’s responsiveness and utility.

AI in Business and Corporate Sector

> 1X Technologies, a Norwegian robotics startup backed by OpenAI and other prominent names, has secured $100 million in Series B funding, led by EQT Ventures with contributions from Samsung NEXT and others. This funding will propel the development and market launch of its innovative bipedal android, NEO, aiming to integrate humanoid robots into the workforce to address global labor demands. With its decade-long journey and partnership with OpenAI, 1X Technologies has advanced its technology, notably creating high-torque servo motors and employing AI models for natural language comprehension and task execution. Its robot, Eve , is already operational in various industrial settings, demonstrating its ability to learn and navigate complex environments. The upcoming consumer-focused Neo, designed for household assistance, reflects 1X Technologies’ ambition to blend humanoids into everyday life.

> Luma AI announced a successful Series B funding of $43M led by Andreessen Horowitz. Key developments include the release of Genie 1.0, a text-to-3D model integrated into their iOS app and web browsers, capable of creating detailed 3D objects in under 10 seconds. Luma AI boasts a user base of over two million, with Genie attracting 100,000 users within four weeks of launch. The company utilizes around 3,000 Nvidia A100 GPUs for training new AI models. Luma AI’s valuation is estimated between $2-300 million, and they have expanded their team with significant hires from NVIDIA and Apple.

> MMGuardian, a company recognized for its advancements in child safety technology, announced the launch of an AI-powered smartphone for kids at CES 2024 in Las Vegas. This innovative phone, developed in partnership with Samsung, uses deep learning models to detect inappropriate content, such as sexting or cyberbullying, thereby alerting parents and children to potential online threats. MMGuardian, established in 2012 and recognized for its contributions to child safety technology, has evolved its capabilities over the years. It introduced scanning for pictures, texts, and chat messages in 2018, leading to over 10 million alerts issued to parents.

> At a Senate hearing on AI and journalism, Congress discussed making tech companies like OpenAI pay for news content used in AI training. Both parties agreed on the need for licensing, with Senators Richard Blumenthal and Josh Hawley leading the discussion. Media leaders advocated for compensation, arguing AI use of their content without licensing is similar to theft. Journalism professor Jeff Jarvis countered, supporting fair use and opposing compulsory licensing. The hearing reflected growing concerns about AI’s impact on journalism and the broader media ecosystem.

> Tennessee Governor Bill Lee announced the Ensuring Likeness Voice and Image Security (ELVIS) Act, a bill designed to protect the music industry from the misuse of AI. The act updates Tennessee’s Protection of Personal Rights law, extending safeguard to songwriters, performers, and music industry professionals against unauthorized AI cloning of their voice. The ELVIS Act is a pioneering step in state-level AI legislation, focusing on the voice’s protection and intended to set a precedent for other states. The bill has garnered support from various industry groups and leaders.

> Swarovski Optik’s innovative AX Visio binocular showcased at CES 2024, offers a unique blend of traditional high-quality optics and advanced AI technology. These smart binoculars, designed by Marc Newson, are equipped with an integrated operating and object recognition system, enabling them to identify over 9,000 species of birds and mammals. Priced at $4,799, it features a 10 x 32 magnification with a 1,000-yard field of view, and the ability to take 13MP photos and 1080p videos, enhancing the wildlife watching experience. It also comes with built-in GPS and a “Share discoveries” function, allowing users to guide others to their spotted wildlife. AX Visio is scheduled for market release in February.

5 new AI-powered tools from around the web

Potis AI is a SaaS tool automating behavioral interviews using real-case examples and AI-powered candidate testing.

Fliz AI is a platform utilizing AI to transform URLs into engaging videos, benefiting e-commerce, real-estate, classified ads, and bloggers. It simplifies content creation and enhances online presence.

Rimo is an AI article generator that transforms audio into text, designed for writers. It blends AI efficiency with human insight, ensuring unique content creation. Beta version is free in January 2024.

kAI is a productivity tool that enhances time management with AI planning, block scheduling, synchronization, custom notifications, and integrates with major apps.

Pre.dev offers instant AI-driven software planning and rapid development, ideal for founders, managers, and agencies. This platform simplifies development stages, provides detailed architecture plans, integrates with tools like Jira, and scales for individuals and enterprises.

Prompt Whisperer simplifies AI interactions with a tool for structuring prompts, ensuring consistency in corporate settings. It optimizes ChatGPT interactions and invites community feedback.

Olly is an AI-powered Chrome extension that generates AI-driven comments, evaluates viral scores, and boosts online presence with a pay-as-you-use model via OpenAI API key.

GPT Auth provides secure authentication, analytics, and payment solutions for GPT applications on the GPT Store. It simplifies user access, usage tracking, and payment processes, offering robust security, detailed analytics, and smooth integration.

arXiv is a free online library where researchers share pre-publication papers.

This paper addresses the disparity in online information access due to language barriers. It introduces over 425K aligned animated video segments in Japanese and English to support automated dubbing, simultaneous translation, guided video summarization, and genre/theme/style classification. This dataset, significantly larger than existing ones, is designed to overcome challenges in automated dubbing like timing, facial movement synchronization, and prosody matching, thus enhancing video accessibility globally. Available at https://github.com/davidmchan/Anim400K, it also aids in various secondary video tasks, with rich metadata for comprehensive analysis and research.

The paper introduces a new approach for inserting 3D objects into scenes using text descriptions and 2D bounding boxes. This method, developed at ETH Zurich and Google Zurich, overcomes the limitations of existing models like Instruct-NeRF2NeRF in placing objects at arbitrary locations with multiview consistency. It involves editing a 2D reference view, lifting this to 3D, and integrating it into the scene, guided by depth estimation. InseRF shows superior results in 3D consistency and control over object placement without needing explicit 3D data, marking a significant advancement in 3D scene editing.

The paper introduces Object-Centric Diffusion (OCD), a novel approach for efficient video editing using diffusion models. Traditional diffusion-based video editing, while producing high-quality results, faces challenges with high memory and computational costs. OCD addresses these inefficiencies with two key techniques: Object-Centric Sampling and Object-Centric 3D Token Merging. Object-Centric Sampling focuses computational resources on more perceptually significant foreground regions, reducing steps in the diffusion process for background areas. Object-Centric 3D Token Merging decreases cross-frame attention costs by merging redundant background tokens. These methods are applied without retraining existing video editing models, drastically reducing memory and computational costs. Extensive evaluations show up to 10× latency reduction while maintaining comparable synthesis quality.

The study introduces AMIE (Articulate Medical Intelligence Explorer), an AI system optimized for clinical history-taking and diagnostic dialogue. AMIE, based on a Large Language Model (LLM), was trained using real-world datasets and a novel self-play simulated dialogue environment with automated feedback. It enhances its diagnostic dialogue capabilities across various medical contexts and specialties. The system’s effectiveness was tested in a randomized, double-blind crossover study with 149 clinical scenarios and 20 primary care physicians (PCPs). The study, resembling an Objective Structured Clinical Examination (OSCE), involved text-based consultations with patient actors. AMIE outperformed PCPs in diagnostic accuracy and was rated superior in 28 of 32 axes by specialist physicians and 24 of 26 axes by patient actors.

This study presents an efficient inference solution for Large Language Models (LLMs) on Intel® GPUs, addressing the challenge of high memory consumption and operational complexity in LLMs. The solution includes simplifying the LLM decoder layer, reducing memory access, and introducing a segment KV cache policy for efficient memory management. These optimizations lead to significant improvements in system latency and throughput. The customized Scaled-Dot-Product-Attention kernel aligns with the fusion policy and KV cache solution. The proposed solution, implemented on Intel® GPUs, showcases up to 7x lower token latency and 27x higher throughput compared to the standard HuggingFace implementation. This advancement significantly enhances the performance of popular LLMs on Intel® hardware.

ChatGPT Creates Comics

Comics drawn and written by ChatGPT + DALLE 3

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.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, apply here.