šŸ¦¾ A Huge Week for AI

On Stability Video Diffusion, Claude 2.1 and 200k Tokens, Scale <> NVIDIA, and Inflection 2

Author

Lior
November 27, 2023

AlphaSignal

Hey ,

Welcome to this week's edition of AlphaSignal.

Whether you are a researcher, engineer, developer, or data scientist, our summaries ensure you're always up-to-date with the latest breakthroughs in AI.

Let's get into it!

Lior

In Todayā€™s Email:

  • Top Releases and Announcements in the Industry

  • Stability Unveils Stable Video Diffusion

  • Anthropic Releases Claude 2.1

  • OpenAI Regains Stability

Read Time: 4 min 43 sec

RELEASES & ANNOUNCEMENTS

Inflection AIā€™s new language model, Inflection-2 outperforms competitors like Google's PaLM 2 on key benchmarks and demonstrates enhanced reasoning and factual accuracy. The model will soon power the Pi chatbot, offering faster and more cost-effective performance.

Amazon announces 'AI Ready,' a commitment to provide free AI skills training to 2 million people by 2025. The initiative includes eight new free AI courses, $12M in generative AI scholarships, and a collaboration with code.org to make AI education accessible.

Karpathy recently uploaded what could be the best non-technical intro to Large Language Models. The 1 hour lecture covers training, inference, jailbreaks, finetuning, the emerging LLM OS and LLM Security.

TensorFlow Hub has integrated with Kaggle Models, offering a wider model collection and enhanced user features. Alongside this, TensorFlow 2.15 has been launched, bringing improvements like simplified CUDA library installation for Linux, oneDNN CPU optimizations for Windows and complete tf.function types availability.

With the Release Preview version, Windows 10 users can now try Microsoft's Copilot AI feature, previously exclusive to Windows 11.

Scale Teams up with NVIDIA to Power the Next Generation of LLMs with NeMo SteerLM

Scale is proud to collaborate with NVIDIA to power the next generation of LLMs for generative AI!

Our collaboration combines Scaleā€™s high-quality training data and NVIDIA NeMo SteerLM: an LLM alignment technique that allows dynamic steering of models during inference. Developers will be able to create applications for enterprise use cases including:

Education: Customize LLM responses based on the personalized needs and preferences of teachers and students.

Gaming: Generate more interactive and immersive non-playable character (NPC) experiences.

Retail: Utilize the chat model to power AI chatbots that generate dynamic responses tuned for various groups of users.

To support generative AI advancements, NVIDIA and Scale have open-sourced the SteerLM dataset. With 37K expert-evaluated samples, it enables developers to build better language models and continue innovating research.

Together, NVIDIA and Scale are excited by the immense potential of combining the latest advancements in generative AI with expert-curated datasets.

NEWS
Stability Unveils Stable Video Diffusion for AI Video Generation

What's New?
Stability AI introduces Stable Video Diffusion, a generative AI video model based on their Stable Diffusion image model. This new tool, available on GitHub and Hugging Face, offers two versions for generating videos with adjustable frame rates. It includes advanced features like text-to-video and image-to-video transformations and has been developed through a multi-phase training process.

Why Does It Matter?
Stable Video Diffusion's flexibility makes it suitable for various video applications like multi-view synthesis from single im

ages. It reportedly outperforms commercial competitors in user preference studies, indicating a significant advancement in video generation. Stability AI aims to develop a series of models that build on and extend this base, similar to the ecosystem that has developed around Stable Diffusion.

Main Takeaways

  • Versatile Video Generation: Text-to-video and image-to-video capabilities with customizable frame rates.

  • High Resolution: SVD can generate 14 frames at 576x1024 resolution. Open-Source Availability: Accessible for research on GitHub and Hugging Face.

  • SOTA Performance: Outperforms commercial counterparts in user preference studies.

Webinar: Master Scalable Data Infrastructures for LLMs

Join the latest webinar by SuperAnnotate's CEO, Vahan Petrosyan.

You will learn to build scalable LLM data infrastructures, focusing on efficient quantization algorithms and complex architectures like Tree of Thoughts.

This webinar offers technical insights into enhancing LLM accuracy and efficiency, crucial for real-world applications, accompanied by a Q&A.

NEWS
Anthropic Releases Claude 2.1 with 200k Token Context Length

What's New?
Anthropicā€™s new large language model (LLM) Claude 2.1 sets a new standard with its 200K token context window, approximately equal to 150,000 words or over 500 pages. Claude 2.1 also introduces a 2x reduction in hallucination rates and improves accuracy when handling complex, factual questions and comprehending intricate documents like legal and technical texts.

Why Does It Matter?
The expanded context window in Claude 2.1 allows for the processing of large documents and entire codebases in a single interaction, surpassing the capacity of its predecessor Claude 2.0 and competitors like GPT-4 Turbo. The reduction in hallucination rates and increased accuracy make it a reliable tool for enterprises in critical decision-making processes.

Main Takeaways

  • 200K Token Context Window: Industry-leading large document uploads.

  • Accuracy Enhancements: 2x reduction in hallucination rates and 30% fewer incorrect answers.

  • Tool Use in Beta: The LLM can leverage user-defined tools, search the web, and retrieve info from private knowledge bases.

  • Comprehension & Summarization: Better handling for complex documents, particularly legal and financial texts.

NEWS
OpenAI Regains Stability: Altman and Brockman are Back

What's New?
Sam Altman and Greg Brockman have returned to OpenAI in leadership roles after a brief period of uncertainty and internal conflict within the company's board. This reinstatement coincides with the formation of a new board, featuring Bret Taylor, Larry Summers, and Adam D'Angelo, tasked with expanding and reshaping the company's governance structure.

Why Does It Matter?
The return of Altman and Brockman signals a stabilization in OpenAI's leadership, pivotal for continuing the company's advancements in AI research and development. This change is expected to influence OpenAI's strategic decisions, potentially impacting the company's approach to AI safety, commercialization, and research directions. The 180 also imbues developers building on top of OpenAIā€™s APIs with renewed confidence.

Main Takeaways

  • Leadership Reinstatement: Altman and Brockman's return brings experienced leadership back to OpenAI.

  • Focus on AI Safety and Ethics: Potential for increased emphasis on responsible AI development under the new leadership.

  • Strategic Partnership Impacts: The leadership change may influence OpenAI's collaborations, especially with Microsoft.

  • Influence on AI Research: These developments at OpenAI could have wide-reaching effects on AI research trends and community dynamics.

Read Last Weekā€™s Summaries:

How was todayā€™s email?

Not Great      Good      Amazing

Thank You

Igor Tica is a contributing writer at AlphaSignal and a research engineer at SmartCat, focusing on computer vision. He's actively seeking partnerships in self-supervised and contrastive learning.

Jacob Marks is an editor at AlphaSignal and ML engineer at Voxel51, is recognized as a leading AI voice on Medium and LinkedIn. Formerly at Google X and Samsung, he holds a Ph.D. in Theoretical Physics from Stanford.

Want to promote your company, product, job, or event to 150,000+ AI researchers and engineers? You can reach out here.