• AI Breakfast
  • Posts
  • NVIDIA Showcases Dialogue With AI Video Game Characters

NVIDIA Showcases Dialogue With AI Video Game Characters

Author

AI Breakfast
May 29, 2023

Good morning. It’s Monday, May 29th.

A New York lawyer is facing a court hearing after his law firm utilized ChatGPT for legal research, resulting in the submission of a brief that cited nonexistent legal cases. The lawyer claimed to be unaware that the tool could produce false information.

In today’s email:

  • Humata: The Premier AI Document Analyzer

  • 5 trending AI tools

  • Watch: NVIDIA showcases dialogue with AI video game characters

  • US lawyer admits using AI for case research

  • AI passed an advertising Turing test for the first time

  • UNESCO unveils new AI Roadmap for classrooms

  • PandaGPT: An AI Foundation Model Capable of Instruction-Following Data Across Six Modalities, Without The Need For Explicit Supervision

  • Latest AI research papers

You read. We listen. Share your feedback by replying to this email, or DM us on Twitter.

Attending Ai4 this year? We will be!

Sponsored post

Humata: The Premier AI Document Analyzer

Humata.ai is an AI-driven tool that is designed to help users analyze and understand their documents more efficiently. It offers an impressive range of features that are intended to speed up research, learning, and report creation.

It's like having an intelligent assistant at your disposal, offering instant Q&A, automatically generating summaries of complex technical papers, and creating content for reports and tasks in an instant.

This tool truly understands context, nuance, and detail. And above all, it takes data security seriously, encrypting your documents in secure cloud storage.

Its free version is robust and offers a great starting point with a 60-page limit, while the Pro Plan unlocks even more powerful features, like querying across multiple documents simultaneously.

Having used all of the GPT-PDF analyzers available, Humata stands out to me as the best one.

Thank you for supporting our sponsors

5 new AI-powered tools from around the web

Meshy is a game-changing 3D generative AI production suite tailored for game developers and creative professionals. This advanced AI texturing and modeling tool is a boon to 3D content generation with its speed and efficiency.

Enhancing knowledge acquisition, Tyles is a practical research tool that extracts valuable insights from diverse sources in a click, acting as your personal assistant for automatic organization and analysis of findings.

An AI-powered video localization tool, Voxqube offers affordable and efficient translation services, trusted by industry giants like Netflix, Sony, and Disney for its impressive expertise.

Mobile Diffusion, an image generator app available on iOS, empowers users to create Stable Diffusion images based on text prompts even without an internet connection, with a priority on privacy.

Equals sets a new standard for spreadsheet versatility with features like real-time collaboration, formulas, charting capabilities, data connectors, calculated columns, and more, making it a comprehensive suite for data manipulation.

A briefing of the latest AI news stories

Nvidia showcased a demo at Computex 2023 that combines gaming and AI, featuring a cyberpunk ramen shop where players can engage in natural speech conversations with video game characters.

The demo, built using Nvidia’s middleware suite called Avatar Cloud Engine (ACE) for Games, allows players to use their own voice to interact with characters.

While the graphical quality is impressive, the dialogue generated by the AI leaves room for improvement. Nvidia plans to release the demo for users to explore different outcomes. The ACE toolkit offers various tools for deploying LLMs and speech-to-text capabilities.

🎧 Did you know AI Breakfast has a podcast read by an actual 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

A New York lawyer is facing a court hearing after his law firm utilized ChatGPT for legal research, resulting in the submission of a brief that cited nonexistent legal cases. The lawyer claimed to be unaware that the tool could produce false information.

AI-generated advertisements have successfully deceived marketing experts and outperformed traditional print ads in a test evaluating creativity and emotional response. In a competition called the “Ad Turing Test”, a panel of 17 marketing experts only achieved 57% accuracy in distinguishing AI-generated ads from human-made ones.

UNESCO convened an important global meeting of Education Ministers to address the opportunities, challenges, and risks posed by generative AI tools in education.

The meeting involved over 40 ministers who shared policy approaches and plans to integrate these tools into education systems. A UNESCO survey revealed that only a small percentage of schools and universities have formal guidance on AI.

UNESCO is developing policy guidelines and competencies framework to guide the use of AI in education, emphasizing the proactive role of teachers in this new era of learning.

PandaGPT is an innovative AI model that combines multimodal encoders and powerful language models to process data across six modalities seamlessly. It has the ability to understand and connect the information from text, image, video, audio, depth, thermal, and IMU inputs.

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

The paper explores the behavior of LLMs in interactive social settings using behavioral game theory. By studying the cooperation and coordination behavior of LLMs, the authors find that LLMs generally excel in games that prioritize self-interest, like the iterated Prisoner’s Dilemma family. However, they struggle in games that require coordination. The analysis focuses on GPT-4 and reveals its unforgiving behavior in the Prisoner’s Dilemma and its inability to coordinate in the Battle of the Sexes.

The paper also suggests methods to modify LLM behavior by providing additional information and predicting the actions of other players. These findings enhance our understanding of LLM’s social behavior and open doors for a behavioral game theory for machines.

This work introduces a closed-loop framework called LLMs As Tool Makers (LATM) that enables large language models (LLMs) to create their own reusable tools for problem-solving. The framework consists of two phases: tool making where an LLM acts as a tool maker and crafts Python utility functions for specific tasks, and tool using, where another LLM acts as the tool user and applies the tools created by the tool maker for problem-solving.

This approach reduces the dependency on existing tools and allows for the cost-effective allocation of tasks between powerful and lightweight LLMs. Experimental results demonstrate the effectiveness of LATM in complex reasoning tasks and achieving performance similar to more resource-intensive models while reducing inference costs.

Backpacks is a groundbreaking computer program that excels in understanding and predicting words in sentences. It introduces a novel neural architecture that combines impressive modeling performance with interpretability and control. It uses special “sense vectors” for each word, which represent different meanings or aspects of the word. We can change these sense vectors to make the program behave differently, and the changes have consistent effects no matter where the word is in the sentence. It performs just as well as other programs, but with Backpacks, we have more insight and control over how it works.

The paper explores the concept of Natural Language-Based Societies of Mind (NLSOMs), inspired by Marvin Minsky’s “society of mind” and Juergen Schmidhuber’s “learning to think.” NLSOMs consist of LLMs and other neural network experts that communicate through a natural language interface. The paper explores how groups of AI models can work together like a society, communicating with each other in a language they understand. They show that these “societies of AI minds” can solve complex problems better than individual models, thereby opening up new possibilities for AI.

In this study, a framework called Incentivized Collaborative Learning (ICL) is proposed to address the challenge of effectively incentivizing entities to collaborate. The framework introduces general design principles for incentives and demonstrates their applicability in federated learning, assisted learning, and multi-armed bandit scenarios. The research aims to achieve win-win collaboration gains and provide modular incentive mechanisms for different use cases, promoting interoperability and practical implementation.

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.

Read by employees from