- AI KATANA
- Posts
- Meta releases two Llama 4 AI models
Meta releases two Llama 4 AI models
Also: Microsoft gives Copilot its own memory in new push to personalize its AI assistant

Hello!
In today’s edition, the AI race continues to heat up as tech giants unveil groundbreaking innovations. Meta has launched two new Llama 4 models—Scout and Maverick—to power its AI assistant across messaging platforms, with a third, even more powerful version still in training. Meanwhile, Microsoft is doubling down on personalization by equipping its Copilot assistant with persistent memory and a suite of new features aimed at building long-term digital relationships. A new study suggests that while AI adoption may initially hinder productivity, those who weather the disruption often come out stronger. Bloomberg sees a future where AI automates the majority of analyst tasks, and DeepMind unveils a detailed framework assessing AI’s potential cyber threats. We also look at major updates to Google Slides powered by Imagen 3, and an eye-popping $40 billion funding round that cements OpenAI’s place in VC history. Let’s dive into the latest in the world of AI innovation, productivity, and risk.
Sliced just for you:
🧠 Meta releases two Llama 4 AI models
📝 Microsoft gives Copilot its own memory in new push to personalize its AI assistant
🦾 If AI doesn’t kill your company, it will make it stronger, study shows
📊 Bloomberg tech chief says AI could streamline 80% of analyst workload
🛡️ Evaluating potential cybersecurity threats of advanced AI
Here’s Why Over 4 Million Professionals Read Morning Brew
Business news explained in plain English
Straight facts, zero fluff, & plenty of puns
100% free
Meta has unveiled two new Llama 4 AI models—Scout and Maverick—designed to power its Meta AI assistant across platforms like WhatsApp, Messenger, Instagram, and the web. Llama 4 Scout is optimized for efficiency, fitting on a single Nvidia H100 GPU while outperforming models like Google’s Gemma 3 and Mistral 3.1. Maverick, Meta’s larger model, rivals OpenAI’s GPT-4o and Google’s Gemini 2.0 Flash, delivering strong performance in coding and reasoning tasks with fewer active parameters. A third model, Llama 4 Behemoth, boasting 288 billion active parameters, is still in training and expected to outperform GPT-4.5 and Claude Sonnet 3.7 on STEM benchmarks. All models employ a “mixture of experts” architecture to boost efficiency. While Meta promotes Llama 4 as open-source, its license restricts commercial use by companies with over 700 million monthly users, raising debates about the true openness of the models. Further developments will be shared at Meta’s upcoming LlamaCon event on April 29.

Microsoft has introduced a major update to its Copilot AI assistant, equipping it with a personalized memory designed to recall user-specific information—ranging from breakfast preferences to life goals—across sessions. Announced during the company’s 50th anniversary event, this feature aims to make interactions more intuitive and adaptive over time. Users can control, manage, or delete what Copilot remembers through a dedicated dashboard, reinforcing Microsoft’s emphasis on user agency. Additional updates include AI agents that perform tasks on users’ behalf, auto-generated podcasts, a deep research assistant, a shopping advisor, and upgraded Bing search capabilities. The move marks Microsoft’s intensified push into personalized consumer AI, positioning Copilot to compete with OpenAI’s ChatGPT memory features, Google’s Gemini, and Amazon’s Alexa+. Microsoft AI CEO Mustafa Suleyman views persistent memory as a cornerstone of building deeper, emotionally resonant AI experiences, signaling a shift from transactional tools toward AI that forms lasting digital relationships.
A new study presented at a European Central Bank conference finds that while early adoption of AI can initially disrupt operations—particularly in manufacturing sectors where automation replaces human labor—companies that endure this short-term turbulence often emerge stronger. Drawing on U.S. Census Bureau data from 2017 to 2021 and a sample of 30,000 firms, researchers discovered that early adopters experienced drops in productivity due to interference with established practices like lean inventory management. However, those that adapted eventually reported gains across sales, productivity, and employment. The rebound was more common in younger, smaller firms, while older, larger companies tended to struggle with the transition. With AI adoption rising from 7.5% to 9.1% during the study period, the findings challenge the assumption that AI always immediately boosts productivity and instead highlight the importance of resilience and adaptability in navigating technological upheaval.
Bloomberg’s chief technology officer, Shawn Edwards, revealed that AI could streamline up to 80% of the workload for financial analysts by automating time-intensive tasks like document review and data extraction. With over 350 specialists focused on AI innovation, Bloomberg is enhancing its Terminal with generative AI tools to unlock insights buried in millions of financial documents, aiming to boost analyst productivity by as much as tenfold in some cases. Edwards emphasized that while human expertise remains vital for asking the right questions, AI provides a clear competitive edge by enabling analysts to rapidly synthesize complex data and develop insights that would previously take hours. This shift marks a broader trend as financial institutions increasingly embrace AI to reduce manual grunt work, especially at junior levels, and compete through better, faster access to actionable intelligence.
A new cybersecurity evaluation framework highlights the dual-use nature of advanced AI systems, outlining how they can both bolster defenses and potentially be weaponized by malicious actors. Drawing from over 12,000 real-world incidents across 20 countries, the framework identifies seven key archetypes of AI-enabled cyberattacks, including phishing, malware, and denial-of-service, while pinpointing stages in the cyberattack chain where AI could reduce costs and increase attack speed and scale. Built on established tools like MITRE ATT&CK but tailored for the AI era, it introduces a 50-challenge benchmark covering everything from reconnaissance to malware development. Early findings suggest that while current AI systems don’t yet offer groundbreaking offensive capabilities in isolation, future frontier models may shift this balance significantly.
🛠️ AI tools updates
Google has rolled out a fresh set of updates to Google Slides, introducing Imagen 3 as its latest AI image generation model to enhance visual storytelling in presentations. Users now have access to a redesigned sidebar housing intuitive design tools, as well as a new library of customizable templates suited for everything from planning to pitching. The updates also include time-saving building blocks—editable elements like quotes, agendas, and key stats—and expanded access to millions of high-quality stock images, GIFs, and web visuals. A new proportional scaling feature allows for precise resizing of grouped content without compromising clarity, marking another step forward in Workspace’s push for smarter, more efficient slide creation.
💵 Venture Capital updates
OpenAI has set a historic milestone in the venture capital world, securing a staggering $40 billion in funding and soaring to a $300 billion valuation—now the second-most valuable VC-backed company after SpaceX. This round, primarily funded by SoftBank with support from Thrive Capital, Coatue, Microsoft, and Altimeter, eclipses all prior private funding rounds, including Ant Group’s $14 billion Series C in 2018. Interestingly, only $10 billion will be disbursed initially, with the remaining $30 billion contingent on OpenAI converting into a for-profit entity by the end of the year. This record-breaking raise underscores the continued surge in generative AI’s appeal and positions OpenAI ahead of ByteDance in the hierarchy of global startup valuations.
🫡 Meme of the day

⭐️ Generative AI image of the day

Before you go, check out Microsoft has created an AI-generated version of Quake
