Sam Altman Joins Microsoft

Author

AI Breakfast
November 20, 2023

Good morning. It’s Monday, November 20th.

Did you know: Speaking of Microsoft, Windows 1.0 was released 38 years ago today.

In today’s email:

  • OpenAI Leadership and Developments

  • AI Ethics and Regulation

  • Corporate Strategy in AI

  • 5 New AI Tools

  • Latest AI Research Papers

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

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

Today’s trending AI news stories

Microsoft CEO hires Sam Altman

“We look forward to getting to know Emmett Shear and [OpenAI’s] new leadership team and working with them.

And we’re extremely excited to share the news that Sam Altman and Greg Brockman, together with colleagues, will be joining Microsoft to lead a new advanced AI research team.”

-Satya Nadella, Microsoft CEO

Over the weekend, Sam Altman was ousted as OpenAI's CEO amid internal tensions and ideological conflicts within the company. OpenAI, originally a research-focused non-profit, evolved into a commercial entity, straining its balance between innovation and responsible AI development. The discord intensified post-ChatGPT release, as it spurred commercial success and operational pressures.

After Altman’s dismissal, key contributors Jakub Pachocki, Aleksander Madry, and Szymon Sidor left the company. Greg Brockman, President and co-founder, also announced his departure. Amidst the turmoil, Ilya Sutskever addressed concerns about the firing being a 'coup', emphasizing the board's commitment to OpenAI's mission for beneficial AGI.

Efforts to reinstate Sam Altman as CEO of OpenAI failed, leading to the appointment of former Twitch CEO Emmett Shear as interim CEO. This decision follows a tumultuous weekend, marked by a failed 5PM PT deadlines for board members to resign and reinstate Altman, triggering employee support for Altman on social media. OpenAI’s direction remains uncertain amid this leadership upheaval.

Sam Altman, prior to his exit was actively fundraising in the Middle East for a new venture, Tigris. This project aimed to establish an AI-focused chip company to compete with Nvidia in the semiconductor market. The venture was still in its early stages of formation and discussions with investors for an ambitious endeavor in the AI and semiconductor industry.

At 3am EDT, Microsoft CEO Satya Nadella posted on X that he has hired Altman and Brockman to run Microsoft’s new Advanced AI Research Lab, and acknowledged Emmet Shear (former CEO of Twitch) as the new OpenAI CEO.

Additionally, OpenAI researcher and board member Ilya Sutskever, who was largely blamed for the ousting, posted “I deeply regret my participation in the board's actions. I never intended to harm OpenAI. I love everything we've built together and I will do everything I can to reunite the company.”

In partnership with CODERABBIT AI

Accelerate Your Code Reviews with CodeRabbit AI

CodeRabbit is here to revolutionize your code reviews with its AI-driven platform. With privacy-focused, contextual pull request reviews, CodeRabbit offers line-by-line code suggestions and interactive chat features to make your coding process more efficient and error-free.

Key Features:

Pull Request Summaries: Understand the intent behind changes with clear summaries and automated release notes.

Line-by-Line Code Suggestions: Receive detailed, actionable suggestions for every line of code changed.

Interactive AI Chat: Engage in contextual conversations within your code lines for better coding solutions.

Customizable Reviews: Tailor the AI to suit your specific coding preferences and needs.

Thank you for supporting our sponsors!

AI Ethics and Regulation

> Germany, France, and Italy have reached a consensus on AI regulation, supporting “mandatory self-regulation through codes of conduct” for AI foundation models, while opposing untested norms. The proposal suggests model cards for AI developers, providing transparency about AI model functions and limitations. This joint agreement is expected to influence broader European AI regulatory negotiations, emphasizing the regulation of AI applications over the technology itself.

> Meta has dissolved its Responsible AI team, which played a crucial role in overseeing ethical AI development. This move has sparked concerns about Meta’s dedication in addressing ethical and safety issues related to AI. The decision comes amid growing scrutiny of AI ethics in the tech sector.

> Open Empathic Project aims to make AI more emotionally intelligent - and you can help. Nonprofit LAION has introduced an open-source initiative focused on imbuing AI systems with empathy and emotional intelligence. The project aims to change the way AI interacts with humans across various domains. LAION is collaborating with volunteers to develop an open-source audio dataset for training empathetic speech synthesis models in AI.

Corporate Strategy in AI

> Amazon is cutting ‘several hundred’ positions from its Alexa voice assistant as it shifts internal resources toward enhancing Alexa’s generative AI capabilities. The affected employees, located in the U.S., Canada, India, and other countries, will receive separation payments and support for job placement. Amazon recently introduced the Let’s Chat feature for Alexa, which removes the need for a wake word before each request. The move suggests Amazon is focusing on improving Alexa’s conversational AI capabilities and may prioritize integration with other applications, similar to Google Assistant’s recent development.

> China has created the National Data Administration (NDA) to oversee its burgeoning data resources and further its digital economy ambitions. The NDA, under the National Development and Reform Commission, will assume responsibilities from the Cyberspace Administration of China. Its objectives include establishing unified data sharing standards, supporting public service digitalization, and coordinating government departments. The NDA’s role in centralizing data control and promoting cross-border data circulation is significant for foreign businesses operating in China.

> Billionaires Xavier Niel, Rodolphe Saadé, and former Google CEO Eric Schmidt are backing a new non-profit AI research lab called “kyutai” in Paris. With a total funding of €300 million, the lab wants to produce open-source research to help France compete in the global AI race. France seeks to develop sovereign AI technology and catch up with countries investing heavily in AI, such as the United States.

5 new AI-powered tools from around the web

Athenic AI is an AI-driven tool for quick, user-friendly data analysis, connecting to Google Sheets and SQL databases. Designed to make data accessible for non-analysts and enhance productivity for data teams, with features like AI training for custom terms and dashboard sharing.

ChatDesigner.ai combines ChatGPT’s conversational AI with Photoshop-like capabilities for image creation and editing. Offers an intuitive chat-based UI for manipulating images, creating portraits, generating product photos, and more.

Cypher offers a unique experience where you can create and interact with AI personas that mimic your own voice and speech patterns. It enables engaging conversations with AI versions of yourself, offering a free platform to explore AI-based communication.

BoostBot streamlines influencer marketing, offering creator recommendations and automated, personalized email communication to influences. It reduces time and effort in influencer outreach, catering to global brands with a huge database of over 274 million influencers.

Linfo.ai simplifies managing digital content by summarizing articles, PDFs, and videos into structured insights. It evolves to autonomously expand and organize information based on user preferences, offering interactive and personalized knowledge management experience.

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

MetaDreamer is an advanced text-to-3D generation tool that efficiently creates high-quality 3D models from text prompts in just 20 minutes. It uniquely separates geometry and texture processing into two stages, leveraging both 2D and 3D prior knowledge. This approach overcomes issues like geometric inconsistency and texture entanglement, making it a leading solution in quick and accurate 3D content generation.

 

The study explores replacing the attention mechanism in Transformers with shallow feed-forward networks, trained via knowledge distillation. Focusing on language transition tasks and using the IWSLT2017 dataset, various replacement methods were tested for encoder and decoder attention layers. The results indicate that these “attentionless Transformers” can perform comparably to the original architecture, but at the cost of increased parameters and reduced flexibility. The finding suggests potential for simpler architectures in complex tasks, pending advances in optimization techniques.

This paper outlines a framework for safely testing Language Model Agents (LMAs) in real-world environments, addressing the challenges of unknown risks and dynamic changes in agents and operational settings. A safety monitor, using model-based supervision, halts agent activities that violate safety boundaries. Tested on AutoGPT, the framework showcases the potential and limitations of ensuring safety in autonomous agent tests, highlighting the need for advanced techniques to manage evolving LMA capabilities and environmental complexities.

UnifiedGPT is a groundbreaking framework that integrates advanced computer vision models with large language models (LLMs), streamlining vision-oriented AI tasks. It automates the workflow from vision pre-processing to post-processing, interpreting user requests through natural language and converting them into visual processing tasks. The framework uniquely combines models like YOLO and SAM with LLMs for adaptive, efficient, and versatile multimodal AI applications, though it faces challenges in keeping pace with rapidly evolving models and ensuring smooth integration.

UFOGen introduces a novel approach to ultra-fast, one-step text-to-image generation. It combines diffusion models with a GAN objective, diverging from traditional methods that focus on samplers or distillation techniques. This hybrid methodology allows UFOFGen to efficiently produce high-quality images from text prompts in a single step. Notably, UFOGen fine-tunes pre-trained diffusion models like Stable Diffusion for this purpose. It stands out for its rapid generation speed, versatility in applications like image-to-image and controllable generation and ability to maintain image quality, marking a significant leap in efficient generative models.

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.