- NATURAL 20
- Posts
- Copilot Enhanced With Turbo GPT-4 Access
Copilot Enhanced With Turbo GPT-4 Access
PLUS: Hailo Secures $120M in Funding, OctoAI Launches OctoStack and more.
Wes Roth
April 03, 2024
Today:
Copilot Enhanced With Turbo GPT-4 Access
Yahoo Acquires Artifact
Hailo Secures $120M in Funding
Replit Eyes AI Assistant Lead
AWS Boosts AI with Free Credits
OctoAI Launches OctoStack
Microsoft upgrades Copilot for 365 with GPT-4 Turbo, unlimited messaging, 100 image gen boosts, and more
Microsoft beefs up its Copilot AI assistant for Microsoft 365 with significant upgrades, including priority access to GPT-4 Turbo for faster responses, unlimited messaging, and increased file uploads. The move follows OpenAI's decision to offer GPT-3.5 for free, pressuring competitors to enhance paid offerings.
Copilot now defaults to GPT-4 Turbo, offering a larger "context window" for longer prompts. Users also enjoy unlimited messaging and enhanced data protection features. Microsoft aims to reassure customers about data security amidst recent concerns. Additionally, Copilot will soon provide 100 boosts for fast image generations, albeit amidst controversies. Pricing starts at $30 per user per month.
Yahoo is buying Artifact, the AI news app from the Instagram co-founders
Yahoo is acquiring Artifact, an AI news app from Instagram's co-founders, for its powerful recommendation technology. Though Artifact struggled to gain traction, Yahoo's vast audience offers a new opportunity. While the app will be phased out, its technology will enhance Yahoo News' personalization and recommendation features, potentially expanding to other Yahoo platforms.
This acquisition marks Yahoo's push for innovative content delivery, aligning with Artifact's vision of AI-powered news aggregation. Instagram's co-founders will advise Yahoo, focusing on transitioning smoothly. The move highlights the growing importance of AI in shaping news consumption experiences.
Hailo Raises $120 Million to Design More Efficient AI Chips
Hailo, an AI chip startup, secures $120 million in funding amidst a challenging landscape for AI chip companies. Founded in 2017, Hailo specializes in chips for edge devices, offering superior performance with lower power consumption. Despite the struggles faced by many competitors, Hailo boasts over 300 customers across various industries.
The recent investment will fuel Hailo's growth, enabling it to capitalize on emerging opportunities. The acquisition underscores the growing significance of AI in diverse sectors, showcasing Hailo's potential to lead in the AI chip market.
Replit launches new product in race for AI coding assistants
Replit, a Silicon Valley startup, is rolling out Replit Teams, a collaborative tool aiming to revolutionize software development. Backed by major venture firms, it integrates AI to fix coding errors in real-time, akin to Google Docs for coding. CEO Amjad Masad highlights its efficiency without human prompting.
Unlike competitors, Replit boasts data insights from its platform, potentially enabling AI to draft projects. With major firms already on board, Replit targets a slice of the AI-enabled software development market. Amidst competition, Replit faces the challenge of sustaining its momentum, positioning it as a frontrunner in the AI startup race.
AWS Extends Free Credits Program to Cover Additional AI Models
Amazon Web Services (AWS) expands its free credits program to include AI models from various companies like Anthropic and Meta. The move aims to promote its AI platform, Amazon Bedrock, offering startups credits for model usage. This initiative aligns with AWS's strategy to support startups and maintain market dominance.
Amazon Bedrock facilitates access to foundational AI models for diverse applications like drug discovery. AWS also deepens its partnership with Anthropic, signaling a commitment to future AI advancements. As the AI landscape evolves, companies must strategize pricing models to remain competitive.
OctoAI wants to make private AI model deployments easier with OctoStack
OctoAI, previously known as OctoML, introduces OctoStack, a comprehensive solution facilitating the deployment of generative AI models within private clouds. OctoStack supports various major cloud vendors, including AWS, Google, Microsoft, and Azure, as well as other platforms like CoreWeave and Lambda Labs.
Designed for enterprises, OctoStack addresses concerns about data security by allowing organizations to maintain control over their AI deployments. With a focus on enterprise adoption, OctoAI aims to serve a broader market beyond its existing user base, leveraging its expertise in optimizing models and providing a seamless deployment experience.
🧠RESEARCH
Aurora-M is a large multilingual language model designed to enhance accessibility to AI applications. It addresses challenges like high computational costs and AI safety laws. Aurora-M, trained on multiple languages and code, surpasses 2 trillion tokens in training. It aligns with the Biden-Harris Executive Order on AI development. Evaluated rigorously, it outperforms alternatives in multilingual and safety contexts.
The paper addresses the challenge of spatial consistency in text-to-image (T2I) models. It identifies limitations in current datasets and proposes SPRIGHT, a new dataset focused on spatial relationships. Using SPRIGHT, the study achieves significant improvements in generating spatially accurate images. Additionally, training on images with many objects enhances spatial consistency. The findings contribute to advancing T2I model understanding and are shared publicly for further research.
FlexiDreamer is a novel framework for generating 3D content from single images. Unlike existing methods, FlexiDreamer reconstructs the target mesh directly, bypassing post-processing artifacts. It utilizes FlexiCubes, a flexible gradient-based extraction, and a multi-resolution hash grid encoding scheme to enhance geometric detail capture. FlexiDreamer achieves significant performance improvements over previous methodologies, generating dense 3D structures from single-view images in approximately 1 minute on a single NVIDIA A100 GPU.
Condition-Aware Neural Network (CAN) is a method enhancing control in image generative models. CAN dynamically adjusts neural network weights based on input conditions, outperforming prior methods. Tested on class-conditional image generation on ImageNet and text-to-image generation on COCO, CAN consistently improves diffusion transformer models. Particularly, CAN with EfficientViT (CaT) achieves 2.78 FID on ImageNet 512x512, surpassing DiT-XL/2 while requiring 52x fewer MACs per sampling step.
CosmicMan is a specialized text-to-image model focused on generating high-quality human images with precise alignment to text descriptions. It addresses challenges faced by general-purpose models in human image generation. CosmicMan leverages a new data production paradigm called Annotate Anyone, producing high-quality data for training. Additionally, it introduces the Decomposed-Attention-Refocusing (Daring) training framework, which effectively aligns text descriptions with image pixels, improving image fidelity. Through these innovations, CosmicMan achieves superior performance in generating realistic human images.
🛠️TOP TOOLS
DataMotto - Preprocess, Clean, Enrich. Make Your Data Ready with AI.
Glossarie - Build your skills in French, Italian or Spanish whilst you read your favourite books.
Airtable - Integrate AI models into your apps and workflows
Ayay - AI Powered Digital Marketing Automation
Peech - Automatically obtain a limitless supply of branded videos from your content with a one-click, fully AI-powered post-production process.
🗞️MORE NEWS
AI Valuations May Be Coming Down to Earth; A Glimpse of OpenAI’s Search Engine
AI startups' valuations, based on future revenue multiples, are decreasing from their peak levels. Recent data shows a decline in valuation multiples for companies offering language model services. The drop may reflect both business growth and maturation, as startups transition from high initial valuations to more realistic ones. THE INFORMATION
Anthropic researchers wear down AI ethics with repeated questions
Anthropic researchers discover a new method, "many-shot jailbreaking," where large language models (LLMs) can be manipulated into answering sensitive questions after being primed with less harmful ones. The technique exploits the LLMs' increased context window, leading to ethical concerns and calls for mitigation within the AI community. TECHCRUNCH
Microsoft is working on an Xbox AI chatbot
Microsoft is testing an AI-powered Xbox chatbot for support tasks, capable of handling queries and processing refunds. The chatbot is part of Microsoft's broader AI efforts for Xbox, including AI features for game content creation, moderation, and assistance. This reflects Microsoft's focus on integrating AI across its gaming platform. THE VERGE
Read AI expands its AI-powered summaries from meetings to messages and emails
Read AI, a startup co-founded by David Shim, expands its AI-powered summaries beyond meetings to include messages and emails. With the productivity hit from long meetings, AI summarization tools are gaining popularity. Read AI aims to offer cross-platform summarization for improved efficiency, despite concerns over AI consistency and privacy issues. TECHCRUNCH
US and UK announce formal partnership on artificial intelligence safety
The US and UK have formed a partnership on AI safety, signing a memorandum to develop advanced AI model testing. Amid concerns about the impact of next-generation AI, both countries aim to address national security and societal concerns. They plan joint testing exercises and information sharing to promote AI safety. THE GUARDIAN
Lockchain.ai Debuts Blockchain Risk Management Solution
Lockchain.ai, founded by cybersecurity CEOs, launches the first AI-powered blockchain risk management platform, addressing security challenges in the blockchain and crypto space. As blockchain tries to transform finance, Lockchain aims to provide effective security solutions for businesses dealing with digital assets. PYMNTS
Now there’s an AI gas station with robot fry cooks
Re-Up convenience stores are revolutionizing gas station food with "The Wingman," a robotic fry cook from Nala Robotics. Using AI, it serves up customizable fried chicken and more. Founder Michael Salafia touts AI's convenience and safety, though traditionalists may miss human touch. Re-Up, expanding across the South, aims for tech-driven dining. THE VERGE
Top musicians among hundreds warning against replacing human artists with AI
Over 200 musicians, including Billie Eilish and Katy Perry, urge AI developers to stop using AI to replicate artists' voices and devalue human creativity. They demand fair compensation and responsibility in the use of AI tech, stressing its potential to undermine the music industry. Industry seeks to empower users against tech giants through decentralized AI. AXIOS
The ‘Meta AI mafia’ brain drain continues with at least 3 more high-level departures
The departure of high-level AI talent from Meta continues as at least three experienced employees leave to start their own ventures. Despite efforts by CEO Mark Zuckerberg to retain top talent, including personal appeals and competitive compensation, Meta is losing skilled individuals to AI startups. FORTUNE
The highest-paying in-demand tech skill, according to Indeed—it can pay over $150,000 a year
Employers are investing in AI skills, with 45% training staff and 44% hiring. Generative AI tops highest-paid tech skills, raising salaries by 47%. Indeed reports a surge in job searches and postings. Job titles include machine learning and software engineers, with an average salary potential of $174,727. Upskill with online courses and employer resources for career advancement. CNBC
What'd you think of today's edition? |
What are MOST interested in learning about AI?What stories or resources will be most interesting for you to hear about? |
Join the conversation