Black Swan - What is the danger of AI?
More topics: Nvidia unveils its vision for AI agents at CES 2025 | Microsoft releases Phi-4 language model on Hugging Face
Hi AI Enthusiasts,
Welcome to this week’s Magic AI News, where we present you the most exciting AI news of the week. Today, we are talking about the potential risks of AI technology and NVIDIA’s announcement at CES 2025.
In addition, we are looking at the new state-of-the-art 14B small language model from Microsoft, which outperforms much larger models on competition math problems. Stay curious! 😎
This week’s Magic AI tool is an AI-powered coding assistant in VSCode based on open-source LLMs. This AI tool can speed up your coding workflow.
Let’s explore this week’s AI news together. 👇🏽
Top AI news of the week
🤔 Black Swan: What is the danger of AI?
Black swans are very rare. A “Black Swan” event is unexpected but has a significant impact. There is also a book on this topic that is worth reading. Politico magazine asked some experts what unpredictable events could occur in 2025. Some see artificial intelligence as a major risk.
The details:
- Gary Marcus (Founder of an artificial intelligence company and author) is concerned about the use of generative AI for cyber attacks. He fears 2025 could see the biggest cyberattack in history, slowing down large parts of the world. Cybercrime is a huge, multi-trillion-dollar problem.
- Cyber attackers can use generative AI to generate well-written texts for phishing attacks via email or search engines. Generative AI tools are also used for coding. Sometimes, programmers do not fully understand the written code, and security vulnerabilities in software products could arise.
- According to Amy Webb (CEO of Future Today Institute), there could be a global stock market crash. AI can analyze real-time data from companies, financial reports, and economic indicators and connect them with public sentiment via social media. Then, a well-planned attack using misinformation on social media could lead to a stock market collapse.
Our thoughts
Disruptive technologies always come with both advantages and disadvantages regarding safety and security. Black Swan events have existed for a long time and will also occur in the future. Predicting them is impossible. For this reason, we should be aware of the risks of AI without panicking. In our view, AI will do more good than harm to humanity.
More information: 🔗 Politico | Cybercrime Magazine
🤖 Nvidia unveils its vision for AI agents at CES 2025
Jensen Huang, CEO of NVIDIA, declared that AI agents are a “multi-trillion-dollar“ opportunity.
Watch the full keynote 👇🏽
The details:
- Nvidia presented a new open platform of world foundation models for physical AI called Cosmos. It is freely available for the physical AI developer community.
- In addition, Nvidia introduced the model families Llama Nemotron and Cosmos Nemotron, which are designed for agentic AI applications.
- Along with the new models, Nvidia also presented blueprints for AI agents. They are designed to help developers create AI applications for specific use cases more quickly.
Our thoughts
In previous newsletters, we’ve mentioned that the AI agent era has started. Nvidia is already preparing for the future by offering frameworks and hardware for AI application development. In the coming years, companies will need to manage swarms of AI agents. Each agent has a different task. We believe that AI agents will shape the future of the economy.
More information: 🔗 Heise Online | NVIDIA
🚀 Microsoft releases Phi-4 language model on Hugging Face
Microsoft published its small language model Phi-4 open-source on Hugging Face. It is the latest member of the Phi family.
The details:
- Phi-4 is a 14B parameter state-of-the-art small language model and excels in complex reasoning in areas such as mathematics.
- The model was trained on high-quality synthetic datasets and is designed for local use.
- It was first released in December on the Azure platform and is now available on Hugging Face and Ollama.
Our thoughts
The new Phi-4 model pushes the boundaries between size and quality. According to Microsoft, it is particularly good at solving math problems and outperforms GPT-4o and Gemini Pro 1.5.
We often use local AI models on our MacBook Pro as coding assistants or for chatting with data. Local models are also a great way to test new software features at a low cost without paying for expensive APIs.
👉🏽 Want to test Phi-4? Check out this week’s Magic AI tool.
More information: 🔗 Microsoft | arXiv
Magic AI tool of the week
Do you ever want a local AI coding assistant? Yes, then our step-by-step guide is what you are looking for! Not a long time ago, most coding assistants were only available via an API of a big provider like OpenAI or Anthropic. Those days are a thing of the past.
Now, you can run state-of-the-art LLMs like Llama 3.2, Phi 4, or Codestral locally on your laptop. The VSCode plugin continue, and the open-source tool Ollama makes it possible.
To learn how to create a local AI coding assistant, check out our blog article. After reading our article, you can chat with your codebase in VSCode and use functions like autocompletion:
👉🏽 Learn more in our article “Mistral’s Codestral: Create a local AI Coding Assistant for VSCode”!
Articles of the week
- Our Journey to Become a Data Scientist
- Build a Local RAG App to Chat with Earnings Reports
- Build an AI Finance Agent Team with phidata
- Build a financial multi-agent system with AG2 (formerly AutoGen) and Ollama
💡 Do you enjoy our content and want to read super-detailed articles about AI? If so, subscribe to our blog and get our popular data science cheat sheets for FREE.
Thanks for reading, and see you next time.
- Tinz Twins
Leave a comment