Navigation auf uzh.ch
Navigation auf uzh.ch
Whether you constantly talk to ChatGPT as if it’s your personal assistant, or you use neural networks and LLMs or other related models directly in your work at UZH, you're almost certainly aware of the issues concerning the future of AI—not just for individuals but also for humanity at large.
As UZH researchers, scientists, scholars, and specialists, it's important to ask the question, "how will I adapt my scholarly work in light of relevant developments in AI (and related technologies)"?
Do you intend to explore (or deepen) your use of more ubiquitous tools such as ChatGPT for coding, drafting text, or otherwise asking general questions? If so, are you aware the parent company of ChatGPT, OpenAI, will soon transition to a more for-profit and private structure?
It’s difficult enough to stay up to date with the technology itself, much less the dealings of organizations, people, and companies behind them. And if these tools are transitioning to the private sphere, you are still left with the questions:
Unsurprisingly (and quite appropriately), we find the answer in our research and scientific principles—especially the imperatives to be open-source and FAIR (Findable, Accessible, Interoperable, and Reusable). This means, in terms of direct actions:
At Science IT, we provide the expertise to guide you to these open-source solutions as well as the university-owned and -operated computational resources to run these tools at scale. In terms of specific expertise in AI, researchers such as Professor Yi-Tang Lin from the Department of History are already working with us. Dr. Lin’s research group is exploring open-source LLMs powered by H2OGPT on Science IT’s powerful in-house GPUs to identify the potential gaps within international organizations’ knowledge on rice planting technologies in West Africa. The group structures “conversational experiments” where they train various LLMs on specific subsets of scientific texts published by international organizations then ask each model the same questions, the process being designed to allow the historians the ability to assess the finer details of how these texts (and their authors) are interrelated and how they contrast when presented with identical queries. Moreover, they can also compare these LLM responses with other historical materials that were not published in their studied organizations’ online databases.
Other professors, such as Dr. Meredith (“Merry”) Schuman from the departments of Geography and Chemistry have also begun working with Science IT to use the same tools to comb through academic literature. Because of the efficiency and speed with which these models can “read” and assess vast amounts of scientific texts, Dr. Schumann has remarked that “these tools could change the process of literature review”. Moreover, with advances such as retrieval augmented generation (RAG), the ability of an LLM to provide responses with direct citations to its sources greatly mitigates the potential for “hallucinations”.
However you intend to make use of AI, LLMs, and other technology in the ever-evolving landscape of scientific research, Science IT can offer guidance, expertise and consulting, as well as the computational power for your work. Deliberately ask yourself:
Get in touch with us via contact@s3it.uzh.ch —we look forward to your inputs!
