João Gonçalves (ESHCC), Sonia de Jager (ESPhil) and Nick Jelicic (Future Library Lab) have received an Academic Grant of 20 000 GPU hours from tech company NVIDIA to train the first generative large language model for the Social Sciences and Humanities (SSH). The team expects that this specialised model will lead to breakthroughs in the social sciences and humanities and allow us to better understand to what extent AI can mimic human reasoning on fields like philosophy, communication and psychology, among others.
The model will be trained exclusively with data sources from the SSH, making it quite different from current state of the art models such as OpenAI’s o1 or DeepSeek-R1, which tend to specialise in fields like mathematics and coding. The ambitious goal of achieving SSH breakthroughs is only made possible through a donation of computational power from NVIDIA, the largest worldwide provider of AI chips. Despite the high volumes of computation involved, Joao Gonçalves remains conscious of sustainability concerns. “The computing power we have been granted is still only a tiny fraction of what commercial language models such as ChatGPT, Grok or Gemini use. In the long term, we also hope to show that smaller, sustainable, specialised language models can perform as well or even better than the larger ones.”
The main goal of this project is research, so it may be able to help with aspects such as historical document analysis, or interpretation of situations from a sociological lens. The discoveries and innovations from an SSH model are likely to result in general language models that are more capable in critical reasoning and knowledge retrieval. The model will be trained in the next six months.
- Researcher
- Researcher
- Researcher
- Related content
