WTS 2024/Submissions/How DBPedia and Wiktionary can make LLMs cheaper and more robust
| ID : | How DBPedia and Wiktionary can make LLMs cheaper and more robust | ||
|---|---|---|---|
| Author(s): Vanshpreet Singh Kohli | Username(s): VKohli17 | Type of submission: Lightning talk | |
| Affiliation: IIIT Hyderabad | Theme(s): Technology | ||
| Abstract:
In this initial phase of what may one day be called an AI revolution, Wikimedia stands to make significant and lasting impact through its efforts in building and sustaining large-scale, publicly available knowledge graphs and dictionaries including DBPedia and Wiktionary, both of which further fuel projects like ConceptNet. Contemporary research demonstrates that leveraging factual knowledge graphs can make LLMs not only more accurate and robust, but also significantly speed up pre-training, saving huge amounts of money and alleviating the carbon footprint of LLMs. |
Slides: PPT Link | ||
| Level of advancement: Advanced | |||
| Special requirements: None | |||
| How will this session be beneficial for the communities ?
Attendees will gain insights into Wikimedia projects like DBPedia and Wiktionary and how they may factor into cutting-edge LLM research. | |||
Interested participants
[edit](register below and ask your questions now to the session organiser)
- ...