April 2025

Show All Agent AI Analytics Computing Leader Marketing Organisation Robots Team

• Microsoft's CEO, Satya Nadella, noted that AI now contributes to 30% of code written at Microsoft.
  | #Leader | Microsoft's CEO, Satya Nadella | 2025/04/30
• Morgan Stanley predicts that humanoid robots will create a $4.7 trillion market by 2050, with about 1 billion units deployed. This growth suggests a major future for humanoid technology, despite concerns over potential challenges that could affect this momentum.
  | #Robots | Morgan Stanley | 2025/04/29
• OpenAI has launched a new image generation model in its API, called gpt-image-1. This model allows developers to create high-quality images directly in their applications. Many companies are already using this technology for various creative tasks, such as logo design and marketing materials. My take : one of the best image generation model is available via API to integrate in marketing flows. Nice !
  | #Analytics | OpenAI | 2025/04/27
• ChatGPT now has the ability to analyze images. This feature helps find any inconsistencies in branding, such as logos, fonts, and colors. It can quickly ensure that new designs match your brand. More examples will be shared in the upcoming newsletter.
  | #AI | digitalsamaritan | 2025/04/22

  @digitalsamaritan

• According to Demis Hassabis, DeepMind's pursuit of artificial general intelligence is accelerating rapidly. With AI now learning like humans and seeing the world visually, it holds potential to cure diseases—though ethical concerns demand urgent discussions on safety, morality, and the future of human-AI coexistence.
  | #AI | Google DeepMind | 2025/04/21
• Google has just released the Gemma 3 model, designed to run efficiently on consumer-grade GPUs like the NVIDIA RTX 3090. Leveraging Quantization Aware Training (QAT), the model significantly reduces memory requirements without sacrificing performance.My take: It looks like Google is actively working toward a future where powerful AI models can be used directly on local machines — no cloud access required.
  | #AI #Computing | Google | 2025/04/18
• Guillaume, Solution Engineer at Salesforce, highlights the company's vision of AI agents as autonomous, reasoning tools that streamline business workflows. He emphasizes that close collaboration between tech and business teams is key to success. Use cases include marketing campaign automation, customer service, complex delivery coordination, and personalized e-commerce recommendations.
  | #AI #Agent #Marketing | Salesforce | 2025/04/16
• OpenAI has released two powerful reasoning models: O3 and O4 Mini. O3 features advanced visual reasoning, enabling image interaction through zoom and rotation, and excels at tasks like extracting handwritten notes and solving visual math problems. O4 Mini is a lightweight, faster version that reduces operational costs while maintaining strong performance. Both models can seamlessly integrate tools such as web search, Python, file analysis, and image generation into their reasoning process. They are the first to "think with images," allowing visual data to be part of their thought chain.
  | #AI #Leader | OpenAI | 2025/04/16

  @aisavvy

• A former Google CEO, Eric Schmidt, predicts that in the next few years, AI will replace many programmers and create advanced systems that can think like humans. This change may lead to superintelligence, which could surpass human understanding and capabilities. He also suggests that recursive self-improvement will significantly enhance AI capabilities, accelerating progress toward general intelligence (AGI), which he expects to be achieved within 3 to 5 years. As the scaling of AI development rapidly advances, the emergence of artificial superintelligence (ASI) — surpassing human intelligence — appears increasingly likely in the future.
  | #AI #Agent #Team | Eric Schmidt | 2025/04/15
• OpenAI has released GPT-4.1, a new model that does better at coding, following instructions, and understanding long texts. It can handle up to 1 million tokens, which helps it work on more complex tasks. You can use this model through the API.
  | #AI | OpenAI | 2025/04/14

  @aisavvy

• Make has introduced new AI agents that leverage GenAI and LLM logic within its automation environment. These agents are easy to set up and can operate in a variety of contexts, such as customer support. They use prompts and tools to remain flexible and guided, allowing them to determine which tasks to perform and how to execute them successfully. Users can create and connect these agents to various tools to explore new use cases. My take: Since Make.com is one of the leaders in web service automation, it will be interesting to test this new feature.
  | #AI #Agent | Make.com | 2025/04/14
• Kling released a new update for their AI video model. Users can now replace or remove elements in videos, achieve better movement and camera control, and benefit from improved prompt coherence. My take: It looks very promising, but if you test it, you'll see that the process isn't intuitive yet. Getting good results requires sometimes many tests and a lot of tweaking.
  | #AI | Kling | 2025/04/14
• At TED 2025, Sam Altman highlighted user feedback as key to OpenAI's progress. He outlined four core principles: improve safety through real-world testing, adapt policies in response to public opinion, prioritize broad user input over elite decision-making, and personalize AI through long-term interaction. Altman said feedback isn't just helpful—it's essential for building safer, smarter, and more user-centered AI systems.
  | #Leader #Organisation | Sam Altman | 2025/04/12
• According to Nvidia Senior Research Scientist Jim Fan, we are close to understanding how to significantly accelerate robot development and enter a new phase of progress. By studying scaling laws, researchers aim to unlock more efficient and intelligent robotics powered by data and advanced technology.
  | #Robots | Nvidia, Jim Fan | 2025/04/09

  @theaigems

• The European Union has unveiled an ambitious plan to establish itself as a global leader in artificial intelligence (AI). As part of this strategy, it will create specialized AI centers—referred to as AI Factories and Gigafactories—equipped with powerful supercomputers. The EU also aims to expand its cloud infrastructure and data storage capabilities, while increasing access to high-quality data for AI development. A comprehensive strategy will support the adoption of AI by businesses and public institutions. Additionally, a dedicated AI skills academy will train the workforce, and initiatives will be launched to attract top international talent. To help smaller companies navigate AI regulations, a support desk will be established. In total, the EU plans to invest €200 billion to drive AI innovation and competitiveness across the continent.
  | #AI #Leader | European Union | 2025/04/09

  @renauddekode

• When developers pass their code to the operations team without proper communication, it causes problems. This "throwing over the fence" approach creates barriers and misunderstandings. Collaboration is key to avoid conflicts and ensure everyone is satisfied with the process.
  | #Analytics | Christophe Blefari | 2025/04/06

  @robindedatagen

• An independent film director, David B., has just unveiled a KFC commercial created entirely with AI — from concept to visuals to voiceover. The ad cost less than $400 to produce, a fraction of the typical budget for traditional commercials. This experiment highlights how AI could revolutionize advertising by drastically reducing production costs and timelines.
  | #Marketing | David Blagojevic | 2025/04/03

  @robonuggets

• Genspark AI has launched the Super Agent, a smart assistant designed to help with complex tasks. According to the company, it can handle things like planning trips, booking restaurants, and creating custom videos to make life easier for everyone. My take: I tried it out, and it delivered some interesting results—especially when I asked it to tackle a more challenging task, like "Find people who work in marketing at large companies in France." As for the more advanced features, such as booking restaurants, I remain skeptical. I tested a few examples, and it seems those services aren't available just yet. But the overall idea—combining many different tools and models into one system to provide users with a full assistant—is definitely intriguing.
  | #Agent | Genspark | 2025/04/02

  @aisavvy

• ChatGPT says it hit 500M weekly users and 20M paid subscribers, up from 15.5M at the end of 2024, which The Information estimates is worth about $415M / mo and $5B / year (based only on $20 / mo tier, so likely higher)—their goal is to hit $12B+ in 2025.
  | #Leader | OpenAI | 2025/04/02
• Kling AI enables fast creation of short product ads using AI. It simplifies video production by automating editing and visuals, making it easy to generate high-quality marketing content in minutes. My take: This will definitely accelerate the production of marketing workflows.
  | #AI | Kling AI | 2025/04/01

  @encyclopedia_ai

• Google, OpenAI, and Anthropic have jointly announced support for the Model Context Protocol (MCP) — an open-source standard developed by Anthropic to securely connect AI models to external data and tools. This protocol represents a major step toward a unified, agent-ready AI infrastructure across the industry. MCP enables seamless, two-way integration between AI systems and real-time applications, allowing models to not only retrieve but also interact with live data and tools. It plays a key role in orchestrating complex, multi-step workflows across various services, making it possible to automate sophisticated tasks through coordinated actions. A major feature of MCP is that it provides access to LLMs with enhanced information retrieval capabilities, including chain-of-thought reasoning, structured context like knowledge graphs, and role metadata. These elements significantly improve the depth, accuracy, and contextual awareness of AI outputs. My take: it's important to look closely at what exactly MCP offers and how to build workflows with it.
  | #AI #Leader | TechCrunch | 2025/04/01
• Coding differs from programming; while coding focuses on syntax, programming requires creativity and problem-solving skills to succeed.
  | #Analytics #Organisation | Fireship | 2025/04/01

  @fireship_dev

• Shubham SHARMA recently shared how he uses AI to make short, personalized videos for LinkedIn to get more replies. My take : it's a smart mix of using automation & AI tools, creating content, and sales skills.
  | #AI #Analytics #Marketing | Shubham SHARMA | 2025/04/01
• Choosing the right AI model depends on the task. Reasoning models like o1-pro, Claude 3.7, and Deepseek r1 are ideal for complex tasks—planning, coding, and brainstorming—with high accuracy but often come at a higher cost and slower speed. GPT-40 stands out as a balanced, everyday model: multimodal, fast, and good at following writing styles. For affordable and efficient automation, Gemini Flash 2.0 and Deepseek v3 excel in data, audio, and video processing. Claude 3.7 also delivers the longest outputs and best one-shot