Hey Synapticians, we have lots to discuss today 😃 

We're excited to explore OpenAI's economic perspective, and we've generated a podcast to help you dive deeper into the document (using NotebookLM, a magical tool that we will showcase later!).

We're also seeing the release of a new model from Mistral, our leading European AI player. It's always encouraging to see them continuing to push the boundaries of European AI model performance!

For this week's theme on "Conversational Search Engines," we've created a short video showcasing Perplexity, a tool that allows you to ask questions in natural language and get responses from an LLM (like ChatGPT) enhanced with real-time internet information. Having verifiable sources for responses is absolutely essential!

Top AI news

1. OpenAI Calls for Strong U.S. AI Investments
OpenAI’s "Economic Blueprint" outlines the need for significant U.S. investment in key AI infrastructure, including semiconductors, data access, and energy resources. These elements are crucial for maintaining a competitive edge in advancing artificial intelligence. The document suggests fostering partnerships between the public and private sectors to fund these initiatives effectively. It also recommends implementing strategic export controls to safeguard sensitive technologies. By prioritizing these actions, OpenAI envisions a sustainable and globally competitive AI ecosystem, ensuring continued innovation and leadership in the field.

2. Mistral AI Releases Advanced Code Generation Model
Mistral AI introduces Codestral 25.01, an advanced AI model designed to revolutionize code generation and developer workflows. Supporting over 80 programming languages, including Python, Java, and C++, the model excels in tasks like fill-in-the-middle (FIM), code correction, and test generation. With an optimized architecture and improved tokenizer, Codestral 25.01 delivers code generation speeds twice as fast as its predecessor. Accessible through IDE plugins for platforms like VS Code and JetBrains, it also offers local deployment options for enterprises. This innovation underscores Mistral AI’s commitment to enhancing coding efficiency and productivity on a global scale.

3. DeepMind Introduces Benchmark to Improve LLM Factuality
Google DeepMind has introduced FACTS Grounding, a benchmark that measures the ability of large language models (LLMs) to generate accurate responses from long-form documents. The initiative aims to address AI hallucinations, where models produce misleading or incorrect information. Accompanied by the FACTS leaderboard on Kaggle, the benchmark ranks models based on their factuality. Google’s Gemini 2.0 Flash leads with an 83.6% score, outperforming models from Anthropic and OpenAI. FACTS Grounding evaluates a wider range of behaviors than previous benchmarks, filling a significant gap in factuality assessment. The leaderboard will be regularly updated to track advancements in AI accuracy.

Bonus. Alexa’s Transformation Faces Big Challenges
Amazon is reimagining Alexa as an AI-powered assistant capable of performing practical tasks, such as suggesting restaurants or adjusting lighting based on user habits. However, the project has hit major hurdles, including addressing "hallucinations" (fabricated AI responses), slow response times, and reliability issues. Rohit Prasad, who leads Amazon’s AI team, emphasized the need for near-zero errors to ensure user trust. Integration with advanced language models and seamless interaction with third-party services has proven more complex than anticipated, delaying the rollout. Despite these challenges, Amazon aims to deliver a more proactive, intelligent, and efficient Alexa experience.

Video of the Day

As you probably already know, OpenAI's competitor, Anthropic, places a strong emphasis on AI alignment. In this video, they delve deeper into what that means for them and how they plan to achieve it!

Theme of the Week

Conversational Search Engines. Practical applications
Discover Perplexity and learn about what is Conversational Search Engines through a real world demonstration!

Stay Connected

Feel free to contact us with any feedback or suggestions—we’d love to hear from you !