Understanding the internal mechanisms of transformer-based language models remains challenging. Mechanistic interpretability based on circuit discovery aims to reverse engineer neural networks by analyzing their internal processes at the level of computational subgraphs. In this paper, we revisit existing gradient-based circuit identification methods and find that their performance is either affected by the zero-gradient problem or saturation effects, where edge attribution scores become insensitive to input changes, resulting in noisy and unreliable attribution evaluations for circuit components. To address the saturation effect, we propose Edge Attribution Patching with GradPath (EAP-GP), EAP-GP introduces an integration path, starting from the input and adaptively following the direction of the difference between the gradients of corrupted and clean inputs to avoid the saturated region. This approach enhances attribution reliability and improves the faithfulness of circuit identification. We evaluate EAP-GP on 6 datasets using GPT-2 Small, GPT-2 Medium, and GPT-2 XL. Experimental results demonstrate that EAP-GP outperforms existing methods in circuit faithfulness, achieving improvements up to 17.7%. Comparisons with manually annotated ground-truth circuits demonstrate that EAP-GP achieves precision and recall comparable to or better than previous approaches, highlighting its effectiveness in identifying accurate circuits.

Life-saving vaccines by could soon be delivered by drone. On-demand drone deliveries of vaccines have several advantages over more traditional means of transport to remote locations and Vanuatu's government has commissioned two companies to test such a programme.

For island nations and countries without the infrastructure for reliable transportation, drones can do more than take photos or collect data: they can transport supplies to save lives. The Pacific island country of Vanuatu, for instance, has teamed up with UNICEF and two drone companies to deliver vaccines to rural areas. Vanuatu is composed of 83 islands spread over an area that covers 1,600 kilometers ( 1,000 miles). To deliver vaccines to its more rural communities, health workers often have to walk for hours -- sometimes, it can even take them days by cars and/or boats. Drones could ensure that local health facilities have quick access to lifesaving supplies when needed.

This September, delivery drones will begin to fly the friendly skies of Vanuatu. And this isn't a one-shot demonstration, like many of the stunts we've seen from the likes of Amazon and Google. This is an attempt to make drones part of the medical infrastructure. The South Pacific island nation of Vanuatu, a string of 83 volcanic islands spread over 1600 kilometers (995 miles), has just issued a "request for tender" to drone companies around the world. The companies are invited to submit bids for bringing vaccines to scattered hospitals and health clinics on three islands.