AI Weekly NewsLetter 11/12/2023

1. Mamba: A New Architecture Beyond Transformers

• This new paper introduces a completely different architecture for Transformers, potentially surpassing its limitations.
• Mamba boasts faster inference (5x faster than Transformers) and the ability to scale up to million-length sequences.
• The paper, code, and models are publicly available for exploration.
• Mamba: Linear-Time Sequence Modeling with Selective State Spaces Paper CodeTry Mamba live here https://huggingface.co/spaces/reach-vb/mamba (Might not be too impressive)

2. Striped Hyena: Another Large Language Model Beyond Transformers

• This 7 billion parameter open-source model offers a glimpse into a world beyond Transformers.
• It features a new architecture that overcomes Transformer bottlenecks, resulting in faster and more memory-efficient training and inference.
• Striped Hyena is available in various sizes for experimentation.
• Paving the way to efficient architectures: StripedHyena-7B, open source models offering a glimpse into a world beyond Transformers Together Launch

3. Mixtral AI’s Mixture of Experts Model

• This 7 billion parameter model is one of the largest AI models ever created.

Read This Tweet to know More Tweet

4. Apple’s New Deep Learning Framework: Nar

• Apple has launched Nar, a deep learning framework specifically optimized for Apple silicon.
• This framework offers pre-built models for various tasks, including linear regression, logistic regression, and Transformers.
• Nar is currently only available for Apple silicon machines.

5. Meta AI’s Purple Llama Initiative

• This initiative aims to promote trust and safety in open-source generative AI models.
• Purple Llama includes a suite of tools and partnerships to address potential risks and biases.
• A workshop will be held next week to discuss the initiative further.
• Llama-Guard : huggingface

6. MagicAnimate AI : A Tool for Animating Images

• This open-source project allows users to upload images and create animations based on dense pose estimation.
• This technology has the potential to revolutionize content creation and entertainment.
• link : Huggingface live

7. OpenML Guide: A Centralized Hub for AI Information

• This community-driven project provides a comprehensive resource for learning about various aspects of AI.
• It covers topics like open-source models, web UIs, and MLOps.
• OpenML Guide is open for contributions from the community.

8. Microsoft Rebrands Bing Chat to Copilot

• Microsoft has rebranded Bing Chat to Copilot, aiming for a more user-friendly name.
• Copilot will integrate features like GPT-4 Turbo and multi-model search.
• This move suggests Microsoft’s continued commitment to AI development.

9. Anthropic Releases New Dataset for Detecting Bias in Language Models

• This dataset includes diverse prompts covering hypothetical situations and sensitive topics.
• It aims to help researchers develop fairer and more ethical AI models.
• dataset link

10. Google’s AI Plans: Gemini and LaMDA

• Google has revealed details about its upcoming AI models, Gemini and LaMDA.
• Gemini is a multi-modal model designed for tasks like code generation and math problem solving.
• LaMDA is a large language model focused on natural language processing and dialogue.
• While both models show promise, their scalability and accessibility remain to be seen.

11. D.D. Sha’s Article on Why Most Agent AIs Don’t Work

• This article explores the limitations of current agent AI architectures.
• It argues that most agent AIs fail to deliver on their promises due to various design flaws.
• The article provides valuable insights for future development of agent AI.

12. OpenAI Power Struggle: Helen Toner Explains the Firing of Sam Altman

• This article sheds light on the recent leadership changes at OpenAI.
• Helen Toner, OpenAI’s CEO, explains the reasons behind Sam Altman’s departure.
• The article raises questions about OpenAI’s internal dynamics and future direction.

13. Meta AI’s Relightable Goian Kodc Avatar

• This project showcases Meta AI’s advancements in creating realistic and animatable avatars.
• The avatars are highly detailed and can be manipulated in real-time.
• This technology has the potential to revolutionize the VR and AR space.
• Link to Learn More

14. Sam Altman Named CEO of the Year by Time Magazine

• This recognition highlights Altman’s contributions to the field of AI.
• However, some argue that the hype surrounding OpenAI and Altman may be excessive.
• Article Link

15. Newly Launched Open models :

1. Mistral 7B MoE Launch Tweet 
2. Magicoder: Source Code Is All You Need https://arxiv.org/pdf/2312.02120.pdf https://github.com/ise-uiuc/magicoder
3. MagicAnimate: Temporally Consistent Human Image Animation using Diffusion Model https://huggingface.co/spaces/zcxu-eric/magicanimate
4. Video to Densepose https://github.com/Flode-Labs/vid2densepose
5. Announcing Purple Llama: Towards open trust and safety in the new world of generative AI https://ai.meta.com/blog/purple-llama-open-trust-safety-generative-ai/
6. Llama-Guard  https://huggingface.co/meta-llama/LlamaGuard-7b  

AI Funding

1. French AI start-up Mistral secures €2bn valuation https://www.ft.com/content/ea29ddf8-91cb-45e8-86a0-f501ab7ad9bb 
2. Replicate raises $40 million Series B led by a16z https://replicate.com/blog/series-b 
3. Announcing our $50M Series C to build superhuman Speech AI models https://www.assemblyai.com/blog/announcing-our-50m-series-c-to-build-superhuman-speech-ai-models/ 

.