diff --git a/_news/locolm-accepted-r2fm.md b/_news/locolm-accepted-r2fm.md new file mode 100644 index 0000000..5e2fc29 --- /dev/null +++ b/_news/locolm-accepted-r2fm.md @@ -0,0 +1,7 @@ +--- +title: "Loco LM @ R2FM" +collection: news +permalink: /news/loco-lm-r2fm +date: 2024-03-06 +--- +How to make LLMs more logically consistent? Check our work at the ICLR 2024 Workshop on on Reliable and Responsible Foundation Models. \ No newline at end of file diff --git a/_publications/calanzone2024locolm.md b/_publications/calanzone2024locolm.md new file mode 100644 index 0000000..239d9c5 --- /dev/null +++ b/_publications/calanzone2024locolm.md @@ -0,0 +1,21 @@ +--- +collection: publications +ref: "calanzone2024locolm" +permalink: "publications/calanzone2024locolm" +title: "Galerkin meets Laplace: Fast uncertainty estimation in neural PDEs" +date: 2024-03-05 10:00 +tags: nesy probml llm +image: "/images/papers/calanzone2024locolm/locolm.png" +authors: "Diego Calanzone, Antonio Vergari, Stefano Teso" +paperurl: "https://openreview.net/forum?id=q3SGbfj19d" +pdf: "https://openreview.net/pdf?id=q3SGbfj19d" +venue: "R2FM Workshop @ ICLR 2024" +excerpt: "We introduce a training objective based on principled probabilistic reasoning that teaches a LLM to be logically consistent with a set of external facts and rules, allowing to extrapolate to unseen but semantically similar factual knowledge." +abstract: "Large language models (LLMs) are a promising venue for natural language understanding and generation tasks. However, current LLMs are far from reliable: they are prone to generate non-factual information and, more crucially, to contradict themselves when prompted to reason about beliefs of the world. These problems are currently addressed with large scale fine-tuning or by delegating consistent reasoning to external tools. In this work, we strive for a middle ground and introduce a training objective based on principled probabilistic reasoning that teaches a LLM to be consistent with external knowledge in the form of a set of facts and rules. Fine-tuning with our loss on a limited set of facts enables our LLMs to be more logically consistent than previous baselines and allows them to extrapolate to unseen but semantically similar factual knowledge more systematically." +bibtex: "@inproceedings{calanzone2024locolm, + title={Galerkin meets Laplace: Fast uncertainty estimation in neural PDEs}, + author={Diego Calanzone, Antonio Vergari, Stefano Teso}, + booktitle={ICLR 2024 Workshop on Reliable and Responsible Foundation Models}, + year={2024} +}" +--- diff --git a/images/papers/calanzone2024locolm/locolm.png b/images/papers/calanzone2024locolm/locolm.png new file mode 100644 index 0000000..79c9401 Binary files /dev/null and b/images/papers/calanzone2024locolm/locolm.png differ