ML6 on Open-Source AI & Hugging Face · Tech & Meet

I attended an engaging Tech & Meet session at Howest presented by Niels Rogge from ML6 and Hugging Face. The session explored how open-source AI is reshaping the technological landscape, and it gave a really clear view of the growing ecosystem around open models, which is the part most people only see in passing on social media.

The framing that mattered

Niels split the AI world into three honest categories instead of the marketing two:

• Closed source. You pay, you call the API, you trust the vendor.
• Open weight. You can run it locally, but you don't get the training data or the recipe.
• Fully open source. Weights, code, and data all in the open.

That distinction matters because it changes the security, control, and customisation story for each. If you're building anything regulated, sensitive, or budget-constrained, the level of "open" you pick decides what you can audit and what you can actually deploy on your own hardware.

He also drew a useful analogy: the Hugging Face ecosystem is becoming for AI what GitHub became for software — the gravitational centre where collaboration, model sharing, and reproducible work all live in one place.

What we covered

• The evolution of large language models — how we got from the first transformers to the current generation, and the practical implications of that pace.
• AI-powered IDEs — the new normal for how developers actually write code, and how that shifts what "tool fluency" means.
• Fine-tuning, inference, and deployment — the operational reality of running models, not just calling them.
• Modern tooling: vLLM, SGLang, and Unsloth AI for serving and training efficiency.
• Hardware and deployment strategy — how to pick the right silicon and serve models efficiently in production instead of over-paying for cloud GPUs.

Why this stuck with me

A few takeaways were directly useful from my angle as a cybersecurity student:

• Open models reduce vendor lock-in, which has real security and resilience implications: you keep the option to inspect, patch, and run things on infrastructure you control.
• AI is now part of the threat model on both sides — attackers and defenders are using it. Knowing the landscape of which models exist and what they're capable of is as important as knowing classical security tooling.
• The HF ecosystem is a learning resource, not just a model registry. Walking through it as a developer is one of the fastest ways to internalise where the field is moving.

What stood out most was how open-source collaboration fosters accessibility, innovation, and transparency in AI. The session had technical depth but also a clear perspective on how open models are driving the next generation of intelligent systems — and how anyone can actually participate.

A big thank you to Niels Rogge and Howest for hosting such an inspiring and insightful evening.