Rohan Deepak Ajwani

Rohan Ajwani, Shashidhar Reddy Javaji, Frank Rudzicz et al. · utoronto

Large Language Models (LLMs) are becoming vital tools that help us solve and understand complex problems by acting as digital assistants. LLMs can generate convincing explanations, even when only given the inputs and outputs of these problems, i.e., in a ``black-box'' approach. However, our research uncovers a hidden risk tied to this approach, which we call *adversarial helpfulness*. This happens when an LLM's explanations make a wrong answer look right, potentially leading people to trust incorrect solutions. In this paper, we show that this issue affects not just humans, but also LLM evaluators. Digging deeper, we identify and examine key persuasive strategies employed by LLMs. Our findings reveal that these models employ strategies such as reframing the questions, expressing an elevated level of confidence, and cherry-picking evidence to paint misleading answers in a credible light. To examine if LLMs are able to navigate complex-structured knowledge when generating adversarially helpful explanations, we create a special task based on navigating through graphs. Most LLMs are not able to find alternative paths along simple graphs, indicating that their misleading explanations aren't produced by only logical deductions using complex knowledge. These findings shed light on the limitations of the black-box explanation setting and allow us to provide advice on the safe usage of LLMs.

Rohan Deepak Ajwani, Zining Zhu, Jonathan Rose et al. · utoronto

Transformer-based Large Language Models (LLMs) have shown exceptional language generation capabilities in response to text-based prompts. However, controlling the direction of generation via textual prompts has been challenging, especially with smaller models. In this work, we explore the use of Prompt Tuning to achieve controlled language generation. Generated text is steered using prompt embeddings, which are trained using a small language model, used as a discriminator. Moreover, we demonstrate that these prompt embeddings can be trained with a very small dataset, with as low as a few hundred training examples. Our method thus offers a data and parameter efficient solution towards controlling language model outputs. We carry out extensive evaluation on four datasets: SST-5 and Yelp (sentiment analysis), GYAFC (formality) and JIGSAW (toxic language). Finally, we demonstrate the efficacy of our method towards mitigating harmful, toxic, and biased text generated by language models.

Rohan Deepak Ajwani, Arshika Lalan, Basabdatta Sen Bhattacharya et al.

We present a Spiking Neural Network (SNN) based Sparse Distributed Memory (SDM) implemented on the Nengo framework. We have based our work on previous work by Furber et al, 2004, implementing SDM using N-of-M codes. As an integral part of the SDM design, we have implemented Correlation Matrix Memory (CMM) using SNN on Nengo. Our SNN implementation uses Leaky Integrate and Fire (LIF) spiking neuron models on Nengo. Our objective is to understand how well SNN-based SDMs perform in comparison to conventional SDMs. Towards this, we have simulated both conventional and SNN-based SDM and CMM on Nengo. We observe that SNN-based models perform similarly as the conventional ones. In order to evaluate the performance of different SNNs, we repeated the experiment using Adaptive-LIF, Spiking Rectified Linear Unit, and Izhikevich models and obtained similar results. We conclude that it is indeed feasible to develop some types of associative memories using spiking neurons whose memory capacity and other features are similar to the performance without SNNs. Finally we have implemented an application where MNIST images, encoded with N-of-M codes, are associated with their labels and stored in the SNN-based SDM.