Hi Synapticians!

Today, the introduction will be entirely dedicated to our latest news — just that 😄

What if we could answer two simple questions:

• Are LLMs actually funny?

• Which LLM is the funniest?

That’s the goal of VibeCheck — a small, fun experiment we just launched. How it works:

• Pick a prompt

• Generate jokes from different LLMs

• Rate them (anonymously)

• See which model makes people laugh (or not)

We believe humor is more than a gimmick — it’s a proxy for emotional intelligence. And as GenAI evolves, EQ is becoming just as important as IQ. Karpathy put it well: "I am prepared to believe humor is really difficult and requires more underlying capability."

NB: at this time Mistral AI is the FUNNIEST :)

NB 2: we will publish all the data on Hugging Face.

Enjoy 😃 

Top AI news

1. Europe unveils bold plan to lead in AI
The European Commission has announced a comprehensive action plan to make Europe the global leader in AI. The strategy focuses on five pillars: computing infrastructure, data access, AI adoption, talent development, and regulatory simplification. Key initiatives include gigafactories with 100,000 AI chips, Data Labs, and the AI Skills Academy. The plan aims to boost AI use in both public and private sectors and attract top global talent. Execution will be key to success.

2. OpenAI launches Pioneers Program to build real-world AI benchmarks
OpenAI has launched the Pioneers Program to develop AI benchmarks tailored to real-world applications in sectors like law, finance, and healthcare. The initiative addresses flaws in current benchmarks, which are often abstract or easy to manipulate. Selected startups will co-create these tools with OpenAI and gain access to reinforcement fine-tuning to improve model performance. The goal is to make AI evaluation more practical and aligned with business realities.

3. Ominimo secures $11M to scale AI insurance
Polish insurtech Ominimo raised €10M from Zurich Insurance at a $220M valuation. Already profitable, it uses advanced AI to analyze over 100 vehicle data points for pricing. With 300,000 policies in Hungary and 7% market share, it plans to expand into 10+ countries. Unlike many insurtechs, Ominimo focuses on backend innovation and real data science, attracting top-tier talent. Zurich becomes both investor and risk carrier, supporting Ominimo’s MGA model. The startup aims to redefine how insurance is priced and delivered, starting with auto and expanding to property.

Bonus. New tool reveals if LLMs memorize or hallucinate
The Allen Institute for AI has released OLMoTrace, a tool that lets users trace language model outputs back to their training data. It highlights exact matches, shows document sources, and distinguishes between memorized, creative, or hallucinated content. The tool is open-source and aims to improve transparency and trust in generative AI. It’s available via the Ai2 Playground and supports several OLMo models.

Tweet of the Day

What do you choose ?

Theme of the Week

AI for Cybersecurity - Paper review

Step into a cyberpunk game of hide-and-seek where AI becomes the ultimate trickster. Discover how scientists are using GenAI to outsmart evolving malware in ways that sound like science fiction—but are very real.

Stay Connected

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