We analyze a family of large language models in such a lightweight manner that can be done on a single GPU. Specifically, we focus on the OPT family of models ranging from 125m to 66b parameters and rely only on whether an FFN neuron is activated or not. First, we find that the early part of the network is sparse and represents many discrete features. Here, many neurons (more than 70% in some layers of the 66b model) are "dead", i.e. they never activate on a large collection of diverse data. At the same time, many of the alive neurons are reserved for discrete features and act as token and n-gram detectors. Interestingly, their corresponding FFN updates not only promote next token candidates as could be expected, but also explicitly focus on removing the information about triggering them tokens, i.e., current input. To the best of our knowledge, this is the first example of mechanisms specialized at removing (rather than adding) information from the residual stream. With scale, models become more sparse in a sense that they have more dead neurons and token detectors. Finally, some neurons are positional: them being activated or not depends largely (or solely) on position and less so (or not at all) on textual data. We find that smaller models have sets of neurons acting as position range indicators while larger models operate in a less explicit manner.

Search Engine Result Pages (SERPs) serve as the digital gateways to the vast expanse of the internet. Past decades have witnessed a surge in research primarily centered on the influence of website ranking on these pages, to determine the click-through rate (CTR). However, during this period, the landscape of SERPs has undergone a dramatic evolution: SERP features, encompassing elements such as knowledge panels, media galleries, FAQs, and more, have emerged as an increasingly prominent facet of these result pages. Our study examines the crucial role of these features, revealing them to be not merely aesthetic components, but strongly influence CTR and the associated behavior of internet users. We demonstrate how these features can significantly modulate web traffic, either amplifying or attenuating it. We dissect these intricate interaction effects leveraging a unique dataset of 67,000 keywords and their respective Google SERPs, spanning over 40 distinct US-based e-commerce domains, generating over 6 million clicks from 24 million views. This cross-website dataset, unprecedented in its scope, enables us to assess the impact of 24 different SERP features on organic CTR. Through an ablation study modeling CTR, we illustrate the incremental predictive power these features hold.

The deep learning era began through the advancements it brought in traditional 2D image-recognition tasks such as classifications, detections, and instance segmentations. As the techniques matured, the research in deep-learning-based computer vision has been shifted towards fundamental 3D computer vision problems -- one of the most notable being synthesising new views of an object and reconstructing the 3D shape of it from images. Many approaches tackled this as a conventional machine learning problem, where the goal becomes to learn a system to "inflate" 3D geometry out of images after a finite set of training iterations. Recently, however, a completely new direction, namely Neural Radiance Fields (NeRF), has been introduced. This article dives into the basic concepts of the originally proposed NeRF as well as several of its extensions in recent years.