Sparse autoencoders (SAEs) and transcoders have become important tools for machine learning interpretability. However, measuring how interpretable they are remains challenging, with weak consensus about which benchmarks to use. Most evaluation procedures start by producing a single-sentence explanation for each latent. These explanations are then evaluated based on how well they enable an LLM to predict the activation of a latent in new contexts. This method makes it difficult to disentangle the explanation generation and evaluation process from the actual interpretability of the latents discovered. In this work, we adapt existing methods to assess the interpretability of sparse coders, with the advantage that they do not require generating natural language explanations as an intermediate step. This enables a more direct and potentially standardized assessment of interpretability. Furthermore, we compare the scores produced by our interpretability metrics with human evaluations across similar tasks and varying setups, offering suggestions for the community on improving the evaluation of these techniques.

Suppose you are alone at home or out for shopping or on vacations and someone breaks into your house. First thing comes into your mind is: if there is some gadget or home security system which can alert you or your neighbors. The home security camera does a good job but they may not work in complete dark. Also, you do not want your gadget to turn on false alarm if it is a cat. In this project I built a proof of concept which merely turn on an LED for demonstration purpose when it detects a person in light or dark using just a Raspberry Pi Pico and a low resolution thermal camera.