OpenAI Registers GPT-5 Trademark

And how Meta is deploying chatbots across platforms

Author

AI Breakfast
August 02, 2023

Good morning. It’s Wednesday, August 2nd.

Did you know: Ambient-temperature superconducting is starting to be replicated after the surprise release of a pre-publication paper explaining how to make the material. Here’s a video on X showing the latest replication.

In today’s email:

  • OpenAI hints at GPT-5

  • Microsoft updates Copilot

  • YouTube trials AI summaries

  • DoNotPay AI bot gains traction

  • AI apps removed in China

  • Nvidia introduces Perfusion

  • Google Ads' AI tool launch

  • AI restores quadriplegic movement

  • Meta's AI chatbots

  • Google enhances Assistant

  • Microsoft warns of chip shortage

  • Together AI's new model

  • 5 New AI Tools

  • Latest AI Research Papers

You read. We listen. Let us know what you think of this edition by replying to this email, or DM us on Twitter.

Today’s edition is brought to you by:

A Banksy got everyday investors 32% returns?

Mm-hmm, sure. So, what’s the catch?

We know it may sound too good to be true. But thousands of investors are already smiling all the way to the bank, thanks to the fine-art investing platform Masterworks.

These results aren’t cherry-picking. This is the whole bushel. Masterworks has built a track record of 15 exits, including net returns of +10.4%, +27.3%, and +35.0%, even while financial markets plummeted.

But art? Really? Okay, skeptics, here are the numbers. Contemporary art prices:

  • outpaced the S&P 500 by 131% over the last 26 years

  • have the lowest correlation to equities of any asset class

Got your attention yet? AI Breakfast readers can skip the waitlist with this exclusive link.

Today’s trending AI news stories

OpenAI Registers GPT-5 Trademark, Hinting at Next-Generation Intelligence in Conversational Models

OpenAI has confirmed the development of GPT-5, its upcoming advanced language model. A recent registration application filed with the US Patent and Trademark Office indicates a potential launch in the near future. Though specific details remain under wraps, OpenAI promises that GPT-5 will represent a significant evolution in natural language processing capabilities.

Microsoft Teases Windows 11 Copilot Integration with Enhanced Third-Party AI Plugins: Microsoft is actively seeking developer feedback to enhance third-party AI plugins for Windows Copilot on Windows 11. This feature aims to centralize AI integration, simplifying user experience and expanding capabilities with Bing and ChatGPT plugins. This development opens new opportunities for developers and a more cohesive user experience.

YouTube Experiments with AI-Generated Video Summaries in Search and Watch Pages: YouTube is trialing an AI-driven feature that auto-generates video summaries for English-language content. The summaries are part of Google's broader exploration of generative AI technology. The initiative's effects on viewer engagement and content creation will become clearer as testing continues.

DoNotPay's AI Lawyer Bot, Breaks Ground in Legal Resolution: DoNotPay, an AI-powered lawyer bot, is transforming the legal industry by successfully overturning parking tickets and resolving related cases. Despite facing legal challenges and criticism over document quality, the service continues to gain traction, offering cost-effective legal assistance.

Generative AI Apps Removed from Apple's China App Store Ahead of Regulatory Changes: In anticipation of new regulations coming into effect on August 15, several generative AI apps have been withdrawn from Apple's China App Store. This action illustrates the ongoing challenge for tech companies to navigate and comply with evolving AI regulations in different regions.

Nvidia's Perfusion: A Compact AI Image Generator with Rapid Training Capability: Nvidia has introduced Perfusion, an innovative text-to-image AI art generator that fits on a floppy disk and trains in just four minutes. With unique features like "Key-locking," Perfusion offers creative flexibility and competitive edge, with plans to release the code soon.

Google Ads Launches AI-Powered, Auto-Generated Advertising Tool: Google Ads' new AI-generated advertisement tool uses generative AI to automatically tailor ads to specific audiences. This move emphasizes the rising influence of AI in the advertising sector, streamlining campaign development and enhancing personalization.

Groundbreaking AI-Enabled Clinical Trial Restores Movement in Quadriplegic Patient: The Feinstein Institutes for Medical Research has utilized a "double neural bypass" system with AI to restore feeling and movement in a quadriplegic individual. This innovative approach offers hope for those with movement impairments and may reshape rehabilitation practices.

Meta's AI-Powered Chatbots Target Enhanced User Engagement: Meta Platforms is reportedly gearing up to introduce a lineup of sophisticated AI-powered chatbots, each with its own unique personality, as early as September. The chatbots are designed to engage users more effectively and enhance the overall user experience on the company’s social media platforms. Meta aims to bolster user retention and maintain its competitive edge in the market by providing new search functionality and personalized recommendations. The move comes amid reports suggesting that Apple is also exploring similar AI-driven offerings.

Google to Elevate Assistant with ChatGPT and Bard-Like Generative AI: Google intends to enhance its Assistant with generative AI features akin to ChatGPT and Google's Bard chatbot. Though exact details remain undisclosed, this move underscores Google's dedication to providing high-quality user experiences.

Microsoft Alerts Potential Service Disruptions Due to AI Chip Shortage: Microsoft has warned of potential service disruptions if it cannot secure enough GPUs for its data centers. The rising demand for AI services has led the company to seek additional capacity and accelerate the deployment of its custom AI processors.

Together AI Releases LLaMA-2-7B-32K: A Long-Context Model for Enhanced Understanding: Together AI has released a 32K context model, LLaMA-2-7B-32K, designed to enhance multi-document understanding and summarization. The model emphasizes the need for open-source options with extended context lengths, encouraging the community to explore further.

🎧 Did you know AI Breakfast has a podcast read by a human? Join AI Breakfast team member Luke (an actual AI researcher!) as he breaks down the week’s AI news, tools, and research: Listen here

5 new AI-powered tools from around the web

Relay is an AI-driven portfolio analyzer that extracts vital information from investment documents and company updates. Users can forward emails to a Relay email address, which automatically converts important data into structured information. It offers historical updates, a portfolio tracking dashboard, team permissions, and access.

Hireguide is an AI-powered platform that enables efficient and equitable interviews by providing structured interview creation and automated note-taking. Founded by LinkedIn alumni, it aims to improve the hiring process and candidate experiences, making it easier for talent teams to conduct effective interviews.

MyLikenessAI is a likeness licensing platform for generative AI that enables artists to monetize and control their likeness in AI-generated content. Generative AI companies can integrate the engine for free, while businesses gain confidence in using AI likenesses commercially with proper licensing and legal protection.

Abridge helps physicians structure, summarize, and get insights on any healthcare conversation. It helps reduce doctor burnout by automatically summarizing conversations into clinical documentation and by integrating with major EHRs. It also has a consumer app to help patients stay on top of their health.

Intelgic offers an Invoice Processing AI model for Accounts Payable automation. With high accuracy and quick processing, it integrates with existing accounting systems, reducing costs, and manual tasks. The AI model supports various document types and ensures data security and compliance.

arXiv is a free online library where scientists share their research papers before they are published. Here are the top AI papers for today.

Skeleton-of-Thought (SoT) is a method introduced to reduce generation latency in large language models. By guiding LLMs to generate a concise skeleton of the answer first and then completing the contents in parallel using API calls or batched decoding, SoT achieves significant speed-up and potential improvements in answer quality. The approach raises questions about organizing LLMs’ thinking in a Graph-of-Thought and explores data-centric optimization for efficiency. While further evaluation and fine-tuning are needed, SoT presents an enticing concept to enhance both efficiency and quality in LLMs by rethinking the necessity of fully sequential decoding and promoting a more human-like thinking process.

The paper presents UnIVAL, a unified model capable of supporting image, video, audio, and language tasks. Unlike other large models, UnIVAL achieves this without relying on massive datasets or billions of parameters. It demonstrates competitive performance and strong generalization capabilities across various image and video-text tasks. The model is efficiently pretrained on many tasks, utilizing multimodal curriculum learning. UnIVAL’s unified approach allows it to achieve competitive results on audio-text tasks, even without audio pretraining. The study also explores multimodal model merging via weight interpolation and highlights the benefits of unification for knowledge transfer and collaboration between tasks. The authors envision UnIVAL as a step towards building modality-agnostic generalist assistant agents.

LLM-Rec explores various promoting strategies to enhance personalized content recommendation performance using Large Language Models and input augmentation. The proposed approach includes four prompting strategies: basic prompting, recommendation-driven prompting, engagement-guided prompting, and a combination of both. Empirical experiments demonstrate that combining the augmented input text with original content descriptions through these prompting strategies leads to improved recommendation performance. The study highlights the significance of incorporating diverse prompts and input augmentation techniques to leverage LLMs effectively for personalized content recommendation. The findings present valuable insights for leveraging LLMs in content recommendation and enhancing user experiences in various domains.

Tencent AI Lab researchers have made strides in enhancing the compositional generalization capabilities of large language models through a novel prompting strategy called “Skills-in-Context” (SKiC). By demonstrating how to compose basic skills to tackle complex problems within the same prompt, SKiC empowers LLMs to achieve near-perfect generalization on challenging tasks. Unlike previous prompting methods, SKiC is a one-stage approach, making it easy to integrate with existing models. The results reveal significant improvements in tasks such as symbolic manipulation, arithmetic operations, and long-context question answering. This advancement could bridge the gap in LLMs’ reasoning capabilities and unlock their latent potential.

This paper introduces ToolLLM, a framework for facilitating large language models to master real-world APIs. It presents ToolBench, a dataset for instruction tuning, which includes 16,000+ real-world APIs from RapidAPI and diverse human instructions involving these APIs. The authors develop a depth-first search-based decision tree (DFSDT) to enhance LLMs’ reasoning capabilities and an automatic evaluator (ToolEval) to assess-tool-use performance. They fine-tune LLaMA on ToolBench, obtaining ToolLLaMA, which demonstrates comparable performance to ChatGPT and robust generalization to unseen APIs. A neural API retriever is developed to recommend relevant APIs for each instruction, making the pipeline more practical.

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.