• AI Emergence
  • Posts
  • OpenAI has made its Safety and Security Committee Independent

OpenAI has made its Safety and Security Committee Independent

Along with: Everything the world is doing with OpenAI-o1

Hey there, 

What is ‘thinking’? 

If you are just recalling information from your memory - would you call it thinking?

Is it different from searching for patterns in everything you have known in the past? And using that to figure out solutions and answers?

To be honest - I don’t know. But that is something I have been thinking about. Not only me - quite a few people are debating whether OpenAI-o1 is thinking or just searching patterns.

What do you think? Is OpenAI-o1 truly thinking?

Think about it, meanwhile, let’s get to this week’s newsletter

What would be the format? Every week, we will break the newsletter into the following sections:

  • The Input - All about recent developments in AI

  • The Tools - Interesting finds and launches

  • The Algorithm - Resources for learning

  • The Output - Our reflection 

Table of Contents

OpenAI established the Safety and Security Committee a few months back in response to growing concerns over the safety of LLM models.

This committee has now been transformed into an independent Board oversight body. It is chaired by Zico Kolter, Director of the Machine Learning Department at Carnegie Mellon University’s School of Computer Science.

The committee also includes notable members such as Quora co-founder and CEO Adam D’Angelo, retired US Army General Paul Nakasone, and former Sony Corporation EVP and General Counsel Nicole Seligman.

Additionally, alongside the full board, it will have oversight of model launches and the authority to delay any release until all safety concerns are fully addressed. (source)

Another shift that’s happening..

Sam Altman, CEO of OpenAI, informed employees in a recent meeting that the company's complex non-profit corporate structure will likely change next year. 

The current structure involves a non-profit controlling a for-profit entity, which has complicated investor relations and growth. 

Altman acknowledged the need for a more traditional for-profit model to streamline operations and meet investor expectations, especially as OpenAI raises new funds. While the non-profit will remain integral, the shift aims to simplify the company's setup and enhance its commercial viability.(source)

YouTube is incorporating new AI-powered tools to enhance creativity, fan engagement, and monetization for creators. The updates are:

  • Integration of Google DeepMind's Veo (Dream Screen) will enable users to create custom video backgrounds and generate 6-second standalone clips for Shorts.

  • The Inspiration Tab in YouTube Studio will be revamped with generative AI to help creators brainstorm video ideas and content like titles and thumbnails.

  • Expansion of its auto-dubbing with a more natural voice for global reach, deeper fan connections via Communities and Hype.(source)

Microsoft recently announced "Wave 2" upgrades for its AI assistant, Copilot, enhancing both enterprise and personal user experiences. The updates include new features like Copilot Pages for collaborative workflows and expanded integration with Microsoft 365 applications. Here are the key highlights:

  • Copilot Pages: Introduces a collaborative workspace for editing and sharing insights within Microsoft 365, focusing on team and group use. Users can create and share Pages with their teams to enhance collaboration.

  • Excel Enhancements: Now generally available with support for formulas, data visualization, and Python integration for advanced analysis using natural language.

  • PowerPoint, Teams, and Outlook: Narrative Builder in PowerPoint is now live; Teams gains meeting and chat synthesis, and Outlook introduces "Prioritize my inbox" for managing important emails.

Additionally, Microsoft introduced Copilot Agents, capable of executing tasks autonomously or with user guidance. Users can build these agents using the new Agent Builder in Copilot Studio, allowing for more personalized and efficient workflows across the Microsoft 365 suite. (source)

A team of technology experts is calling for challenging questions to test AI systems that have easily passed popular benchmarks. This initiative, led by the Center for AI Safety (CAIS) and Scale AI, aims to push AI models beyond their current limits.

The announcement follows the preview of OpenAI’s new model, o1, which "destroyed the most popular reasoning benchmarks,",’ according to CAIS director Dan Hendrycks. To prevent AI from relying on memorized responses, some questions from this ‘Humanity’s Last Exam’ will remain private.

The project seeks at least 1,000 crowd-sourced questions by November 1. These will undergo peer review, with winning entries awarded up to $5,000 and co-authorship, sponsored by Scale AI.

Do you have a question for Humanity’s Last Exam? (source)

Mistral AI has introduced a free tier for developers to test and fine-tune their AI models, including multimodal capabilities such as image processing. This allows developers to explore and build test applications without upfront costs. 

The company also reduced the pricing for its API services and enhanced its consumer AI chatbot, Le Chat, with image recognition features. These changes make Mistral’s AI tools more accessible, especially for developers looking to experiment with AI capabilities affordably. (source)

Salesforce has launched Agentforce, a suite of low-code tools that enable enterprises to build autonomous AI agents for tasks in sales, service, marketing, and commerce. Unlike traditional chatbots, these agents use the Atlas reasoning engine to autonomously perform complex decision-making and workflows. Features include:

  • AI Agents: Automate business processes with human-like reasoning, capable of updating records, analyzing customer sentiment, and optimizing marketing campaigns.

  • Out-of-the-box Agents: Includes service agents, merchant agents, and campaign optimizers to enhance customer interactions.

  • Availability & Pricing: Launches on October 25, 2024, starting at $2 per conversation, with a network of partners like AWS and Google supporting the suite. (source)

Post the release of o1 - OpenAI’s reasoning model. People have shared some interesting use cases -

  • Building an entire chess game -(source)

  • Write PhD code in 1 hour (that took him ~1 year) -(source)

  • Create a full weather app for iOS -(source)

Tweet of the week

Tool: NoteBookLM

Google's experimental NoteBookLM tool simplifies complex information quickly. You can interact with it in various ways, like generating audio overviews, asking questions about documents, and creating FAQs or study guides.

How to Get Started:

  1. Visit NoteBookLM

  2. Click "Try NotebookLM"

  3. Create your first notebook

  4. Add your information source

  5. Explore features like asking questions and generating audio overviews

  • All-in has released podcast videos from the All-in Summit 2024. A lot of prominent figures were part of it - Elon Musk, Sergey Brin, Peter Thiel, and Travis Kalanick. You can check out the playlist here.

  • RAGs have become the preferred method for building enterprise-level applications. You can start your journey to creating RAGs with this free course

What was your answer about OpenAI-o1 thinking?

Now try FiveThirtyNine and tell whether it is thinking or not. This question about what is thinking and what is Intelligence is going to define how we will look at the coming AI tools and how they would augment/complement/replace us.

Are models like OpenAI-o1 and FiveThirtyNine thinking?

Login or Subscribe to participate in polls.

How do you rate this issue of AI Emergence?

Would love to hear your thoughts

Login or Subscribe to participate in polls.

Reply

or to participate.