Add to Chrome
Add to Firefox
Add to EdgeNikhil Prakash
Fine-Tuning Enhances Existing Mechanisms: A Case Study on Entity Tracking
Feb 22, 2024
Fine-tuning on generalized tasks such as instruction following, code generation, and mathematics has been shown to enhance language models' performance on a range of tasks. Nevertheless, explanations of how such fine-tuning influences the internal computations in these models remain elusive. We study how fine-tuning affects the internal mechanisms implemented in language models. As a case study, we explore the property of entity tracking, a crucial facet of language comprehension, where models fine-tuned on mathematics have substantial performance gains. We identify the mechanism that enables entity tracking and show that (i) in both the original model and its fine-tuned versions primarily the same circuit implements entity tracking. In fact, the entity tracking circuit of the original model on the fine-tuned versions performs better than the full original model. (ii) The circuits of all the models implement roughly the same functionality: Entity tracking is performed by tracking the position of the correct entity in both the original model and its fine-tuned versions. (iii) Performance boost in the fine-tuned models is primarily attributed to its improved ability to handle the augmented positional information. To uncover these findings, we employ: Patch Patching, DCM, which automatically detects model components responsible for specific semantics, and CMAP, a new approach for patching activations across models to reveal improved mechanisms. Our findings suggest that fine-tuning enhances, rather than fundamentally alters, the mechanistic operation of the model.
Denoising Heat-inspired Diffusion with Insulators for Collision Free Motion Planning
Oct 19, 2023Diffusion models have risen as a powerful tool in robotics due to their flexibility and multi-modality. While some of these methods effectively address complex problems, they often depend heavily on inference-time obstacle detection and require additional equipment. Addressing these challenges, we present a method that, during inference time, simultaneously generates only reachable goals and plans motions that avoid obstacles, all from a single visual input. Central to our approach is the novel use of a collision-avoiding diffusion kernel for training. Through evaluations against behavior-cloning and classical diffusion models, our framework has proven its robustness. It is particularly effective in multi-modal environments, navigating toward goals and avoiding unreachable ones blocked by obstacles, while ensuring collision avoidance.
Discovering Variable Binding Circuitry with Desiderata
Jul 07, 2023
Recent work has shown that computation in language models may be human-understandable, with successful efforts to localize and intervene on both single-unit features and input-output circuits. Here, we introduce an approach which extends causal mediation experiments to automatically identify model components responsible for performing a specific subtask by solely specifying a set of \textit{desiderata}, or causal attributes of the model components executing that subtask. As a proof of concept, we apply our method to automatically discover shared \textit{variable binding circuitry} in LLaMA-13B, which retrieves variable values for multiple arithmetic tasks. Our method successfully localizes variable binding to only 9 attention heads (of the 1.6k) and one MLP in the final token's residual stream.
Conceptualization and Framework of Hybrid Intelligence Systems
Dec 11, 2020
As artificial intelligence (AI) systems are getting ubiquitous within our society, issues related to its fairness, accountability, and transparency are increasing rapidly. As a result, researchers are integrating humans with AI systems to build robust and reliable hybrid intelligence systems. However, a proper conceptualization of these systems does not underpin this rapid growth. This article provides a precise definition of hybrid intelligence systems as well as explains its relation with other similar concepts through our proposed framework and examples from contemporary literature. The framework breakdowns the relationship between a human and a machine in terms of the degree of coupling and the directive authority of each party. Finally, we argue that all AI systems are hybrid intelligence systems, so human factors need to be examined at every stage of such systems' lifecycle.