75th Data Science Leuven meetup
Details
19h: Doors open.
19h30, talks:
- Thomas Verelst will speak about his work at Axelera.AI making neural nets practical to deploy on custom chips.
- Douw Marx. Does your LLM care about the same things you do? RLHF, reward models and mild AI-safety marketing.
This talk will introduce the basics of LLM post-training, preference models, and the importance of using preference models that match your values. To do this, he will present a side project in which I measured the sensitivities of LLM reward models to different "constitutional principles". How sensitive is the LLM reward model to murder?
He will also include some practical tools for measuring the risks associated with AI models. - Ma..... Tr.....
21h: Drinks at STUKcafé.
🎤 We’re almost on the look-out for interesting speakers and topics. Submit your talk here: https://sessionize.com/data-science-leuven