João Gonçalves has been active across several fronts in recent weeks, spanning national coordination on AI in the humanities, a major compute grant, an innovation challenge, and an upcoming international conference appearance.
On June 5th, Gonçalves organised the Humane AI Sectorplan meeting at Erasmus University Rotterdam, bringing together more than 40 participants from 10 Dutch universities to discuss the national sectorplan through 2028. Discussions covered impact, research, education and dissemination, with a focus on how humanities and social science faculties can best collaborate on AI at the national level. Photos from the event are available via Brigid upon request.
Gonçalves was also awarded a large compute grant from NWO worth 6.7 million SBUs (roughly 35,000 compute hours, or about €110,000) to complete training of the SHARE (Social-Humanities AI for Research and Education) models. The grant supports compute-optimal training of what is intended to be the first generative AI model built specifically for the social sciences and humanities, using the Snellius supercomputer. The model is expected to be fully trained by September 2026.
Last Friday, Gonçalves took part in the Futureproof AI Challenge in Amsterdam, teaming up with representatives from ABN AMRO, UvA, TNO, DNB, UWV, Hogeschool Utrecht and AI-Hub Noord Nederland. The team proposed and pitched a Sovereign AI roadmap aimed at helping organisations take their first steps toward autonomy and control in AI. The pitch impressed the jury, who agreed to back the project for the next six months to develop a minimum viable product.
Looking ahead, Gonçalves will present at the Pluralistic Alignment Workshop, part of the International Conference on Machine Learning (ICML) in Seoul, South Korea, on July 11th. ICML is among the leading conferences in the machine learning field, having attracted more than 30,000 submissions this year. At the workshop, Gonçalves will defend a position paper arguing that mass media regulation frameworks should inform the regulation of post-training alignment datasets for large language models. While in Seoul, Gonçalves will also visit the Center for Trustworthy AI at Seoul National University, a long-standing collaborator.
- Researcher
