arXiv subjects

V. Gurucharan

Publications and source records attributed to V. Gurucharan.

4 recordsLinked to original sources

arXiv2024

Can LLMs faithfully generate their layperson-understandable 'self'?: A Case Study in High-Stakes Domains

Large Language Models (LLMs) have significantly impacted nearly every domain of human knowledge. However, the explainability of these models esp. to laypersons, which are crucial for instilling trust, have been examined through various skeptical lenses. In this paper, we introduce a novel notion of LLM explainability to laypersons, termed $\textit{ReQuesting}$, across three high-priority application domains -- law, health and finance, using multiple state-of-the-art LLMs. The proposed notion exhibits faithful generation of explainable layman-understandable algorithms on multiple tasks through high degree of reproducibility. Furthermore, we observe a notable alignment of the explainable algorithms with intrinsic reasoning of the LLMs.

cs.HC↗

arXiv2024

LaMSUM: Amplifying Voices Against Harassment through LLM Guided Extractive Summarization of User Incident Reports

Citizen reporting platforms help the public and authorities stay informed about sexual harassment incidents. However, the high volume of data shared on these platforms makes reviewing each individual case challenging. Therefore, a summarization algorithm capable of processing and understanding various code-mixed languages is essential. In recent years, Large Language Models (LLMs) have shown exceptional performance in NLP tasks, including summarization. LLMs inherently produce abstractive summaries by paraphrasing the original text, while the generation of extractive summaries - selecting specific subsets from the original text - through LLMs remains largely unexplored. Moreover, LLMs have a limited context window size, restricting the amount of data that can be processed at once. We tackle these challenges by introducing LaMSUM, a novel multi-level framework combining summarization with different voting methods to generate extractive summaries for large collections of incident reports using LLMs. Extensive evaluation using four popular LLMs (Llama, Mistral, Claude and GPT-4o) demonstrates that LaMSUM outperforms state-of-the-art extractive summarization methods. Overall, this work represents one of the first attempts to achieve extractive summarization through LLMs, and is likely to support stakeholders by offering a comprehensive overview and enabling them to develop effective policies to minimize incidents of unwarranted harassment.

cs.CL↗

arXiv2017

Anomalous Dimensions in Non-Supersymmetric Bifundamental Chern-Simons Theories

Non-abelian Chern-Simons theories coupled to fermions are known to provide an interesting class of non-supersymmetric conformal fixed points \cite{Giombi:2011kc}. These theories, particularly those based on bifundamental matter, are important because they may provide simple non-supersymmetric examples of the AdS/CFT correspondence. For instance, it seems natural to conjecture that $O(N)_{-k}\times O(N)_k$ Chern-Simons theory coupled to Majorana fermions transforming in a bi-vector representation may be dual to pure Einstein gravity with a small negative cosmological constant in the "M-theory" limit where $k=1$ and $N$ is large. While it is extremely difficult to directly study such bifundamental theories when $k=1$ or even at strong 't Hooft coupling $λ=\frac{N}{k}$, it is possible to calculate physical quantities to all orders in $λ$ in a $U(M)_{k_M} \times U(N)_{k_N}$ theory, in the limit $M \ll N$, in an $M/N$ expansion. To illustrate this, we calculate the anomalous dimension of the primary operator $\barψψ$, to first order in $M/N$, to all orders in $λ_M=\frac{N}{k_M}$, but with $λ_N=\frac{N}{k_N}=0$. We also comment on possible bosonization dualities for bifundamental Chern-Simons theories.

hep-th↗

arXiv2017

On the Higher-Spin Spectrum in Large N Chern-Simons Vector Models

Chern-Simons gauge theories coupled to massless fundamental scalars or fermions define interesting non-supersymmetric 3d CFTs that possess approximate higher-spin symmetries at large N. In this paper, we compute the scaling dimensions of the higher-spin operators in these models, to leading order in the 1/N expansion and exactly in the 't Hooft coupling. We obtain these results in two independent ways: by using conformal symmetry and the classical equations of motion to fix the structure of the current non-conservation, and by a direct Feynman diagram calculation. The full dependence on the 't Hooft coupling can be restored by using results that follow from the weakly broken higher-spin symmetry. This analysis also allows us to obtain some explicit results for the non-conserved, parity-breaking structures that appear in planar three-point functions of the higher-spin operators. At large spin, we find that the anomalous dimensions grow logarithmically with the spin, in agreement with general expectations. This logarithmic behavior disappears in the strong coupling limit, where the anomalous dimensions turn into those of the critical O(N) or Gross-Neveu models, in agreement with the conjectured 3d bosonization duality.

hep-th↗