Snowflake Arctic 128 Experts MoE

PLUS: Augment New AI Coding Tool, Cognition AI Hits $2 Billion Valuation and more.

Today:

  • Snowflake Cost-Efficient Arctic Model

  • OpenAI Boosts Moderna’s mRNA

  • Synthesia Expressive Avatars

  • Augment New AI Coding Tool

  • Cognition AI Hits $2 Billion Valuation

BIG win for Open Source AI | Snowflake Arctic 128 Experts MoE, "Cookbook" create world-class models

Snowflake has launched Arctic, a large language model tailored for business-to-business applications, emphasizing cost efficiency and specific enterprise needs like SQL generation and coding. Unlike consumer-focused models, Arctic is designed to be less resource-intensive and more accessible, with costs dramatically lower than typical industry standards—under $2 million compared to tens or hundreds of millions. 

It is distinguished by its use of a hybrid architecture that blends dense and sparse model elements, and its open Apache 2.0 license allows free access to model weights and code, encouraging the wider AI community to develop custom solutions affordably. Snowflake's introduction of Arctic into the AI landscape highlights a shift toward open-source models, potentially increasing innovation while lowering costs across the sector.

Moderna and OpenAI Collaborate To Advance mRNA Medicine

Moderna and OpenAI have teamed up to push the boundaries of AI in healthcare. By integrating OpenAI's generative AI into Moderna's operations, they aim to enhance efficiency and impact in mRNA medicine development. Moderna has already integrated ChatGPT Enterprise across its functions, aiding employees with personalized support. 

This collaboration reflects Moderna's commitment to innovation and scalability, crucial for their ambitious product launch plans. With over 750 deployed GPTs driving automation and productivity, Moderna's focus is on optimizing vaccine dose profiles while prioritizing safety. The partnership underscores the transformative potential of AI in shaping the future of healthcare.

Synthesia takes next leap in AI video with ‘expressive avatars’

London-based startup Synthesia has introduced "expressive avatars," a groundbreaking advancement in AI video technology. These avatars dynamically adjust tone, facial expressions, and body language based on context, enhancing realism in AI-generated videos. Unlike Microsoft's VASA, still in research, Synthesia's technology is live, offering customers unprecedented realism in their video content creation. 

Founded in 2017, Synthesia has seen widespread adoption in enterprises, with over 200,000 users creating 18 million videos. The introduction of expressive avatars fills a crucial gap in digital avatars' ability to understand and convey sentiment, promising more engaging and empathetic AI videos. With safety measures in place, Synthesia aims to foster positive, responsible use cases across industries, including healthcare and marketing.

Eric Schmidt-backed Augment, a GitHub Copilot rival, launches out of stealth with $252M

HTML class=”3”

Augment, a startup backed by Eric Schmidt, emerges with $252 million in funding, aiming to revolutionize AI-driven coding. Founded by ex-Microsoft developer Igor Ostrovsky and AI scientist Guy Gur-Ari, Augment boasts a stellar team including Scott Dietzen and Dion Almaer. With industry giants like Microsoft and Google already in the AI coding space, Augment faces fierce competition. 

However, Ostrovsky claims early success with "hundreds" of developers using Augment in early access. The startup aims to address concerns around code quality and intellectual property while accelerating product development with its recent funding boost.

Six-Month-Old AI Coding Startup Valued at $2 Billion by Founders Fund

In just six months, Cognition AI, a startup, soared to a $2 billion valuation with a $175 million investment from Founders Fund. Despite a shaky product launch, the allure of a chatbot capable of coding and debugging attracted hefty investments. Founders Fund had previously led a Series A investment valuing the startup at $350 million. 

The startup's AI-powered coding assistant, Devin, caught investors' attention, propelling Cognition AI to unicorn status. The rapid rise underscores the growing interest and confidence in AI-driven solutions, signaling a potential shift in tech investment trends.

🧠RESEARCH

"PuLID: Pure and Lightning ID Customization via Contrastive Alignment" presents a new method for text-to-image generation called PuLID. It combines Lightning and diffusion branches to enhance ID customization without disturbing the original model. PuLID ensures high ID fidelity and preserves image consistency. Experimentation confirms its superior performance in ID fidelity and editability.

"CatLIP: CLIP-level Visual Recognition Accuracy with 2.7x Faster Pre-training on Web-scale Image-Text Data" introduces a novel method for pre-training vision models using web-scale image-text data. By reframing pre-training as a classification task, it eliminates the need for computationally intensive pairwise similarity computations, achieving a significant 2.7x speedup compared to contrastive learning. 

"ID-Aligner: Enhancing Identity-Preserving Text-to-Image Generation with Reward Feedback Learning" tackles challenges in identity-preserving text-to-image (ID-T2I) generation. Existing methods struggle with maintaining accurate identity characteristics and aesthetic appeal. ID-Aligner introduces feedback learning to address these issues, leveraging rewards from face detection/recognition models and human preferences. 

"Editable Image Elements for Controllable Synthesis" introduces a method to facilitate image editing within diffusion models. By encoding input images into editable "image elements," users can intuitively manipulate them for tasks like resizing, rearranging, and removing objects. The approach enables spatial editing and realistic image reconstruction.

"MotionMaster: Training-free Camera Motion Transfer For Video Generation" addresses the challenge of flexible camera motion control in video generation. Existing methods rely on training complex temporal camera modules, limiting flexibility and requiring significant computation. MotionMaster proposes COMD, a training-free model that disentangles camera and object motions, enabling flexible and diverse camera control. It achieves this through one-shot and few-shot camera motion disentanglement methods, along with motion combination techniques.

🛠️TOP TOOLS

Popstarz - Find the popstar in you

Spline - AI 3D Generation

Airtable - Use AI to analyze, organize, and connect the workflows and information you already have in Airtable.

Pickaxe - Launch a suite of GPTs with your expertise, your brand, your data—your rules.

Rezi - resume platform that uses leading AI to automate every aspect of creating a hirable resume—writing, editing, formatting, and optimizing.

📲SOCIAL MEDIA

🗞️MORE NEWS

Salesforce Einstein Copilot brings new reasoning and actions to enterprise generative AI 

Salesforce’s Einstein Copilot, now widely available, goes beyond basic chatbots by enabling actions through generative AI. It connects to various data sources, enhancing context for better performance. With Copilot Actions, users can trigger workflows, boosting productivity. Advanced reasoning techniques and analytics further optimize enterprise workflows, with ongoing improvements in efficiency and cost-effectiveness. VENTUREBEAT

Toyota and Nissan pair up with Tencent and Baidu for China AI arms race

Toyota and Nissan forge partnerships with Tencent and Baidu in China's AI race at the Beijing Auto Show. Facing challenges like declining EV demand and local competition, Japanese automakers integrate AI tech from Chinese giants to stay relevant and explore smart car development, reflecting China's evolving automotive landscape. CNBC

Oracle adds AI capabilities to its Fusion Cloud CX

Oracle enhances its Fusion Cloud CX with new AI features, including generative AI-assisted answer generation and opportunity quality scoring, aimed at streamlining sales, marketing, and service operations. Competing with Salesforce, ServiceNow, and SAP, Oracle seeks to empower enterprises with smarter customer engagement tools, amid a landscape of evolving CRM offerings. CIO

Perplexity AI: Exploring AI-powered search beyond Google

Perplexity AI challenges Google search dominance with conversational answers and citations. It's powered by ChatGPT 3.5 and aims to reshape search experience by summarizing data and providing cited sources. While still small compared to Google, it's gaining traction, offering potential for brands and marketers to explore new search strategies. SEARCH ENGINE LAND

UK Set to Revise AI Oversight Amid Big Tech Data Boom

The UK plans to revise AI oversight for big tech's data access, aiming to enhance financial products and consumer options. While EU tends to regulate tech pre-market, UK prefers collaboration with big tech. Such moves globally reflect increased scrutiny and potential regulation to balance innovation and consumer protection. PYMNTS

AI designs active pharmaceutical ingredients quickly and easily based on protein structures

Chemists at ETH Zurich develop a groundbreaking AI algorithm that designs active pharmaceutical ingredients matching protein structures. This process streamlines drug discovery by generating molecules that precisely interact with target proteins, potentially revolutionizing pharmaceutical research and development. Tests with industry partners demonstrate the algorithm's efficacy and potential for advancing drug discovery. PHYS.ORG

What'd you think of today's edition?

Login or Subscribe to participate in polls.

What are MOST interested in learning about AI?

What stories or resources will be most interesting for you to hear about?

Login or Subscribe to participate in polls.

Reply

or to participate.