In this work we apply t-sne to the spike sorting process and generate 2D plots that show obvious clusters of spikes. We use two types of data to validate our technique. The first is a ground-truth dataset that comes from paired recordings \cite{neto_validating_2016} with an extracellular and a juxtacellular probe, thus providing labels from the juxtacellularly recorded unit within the extracellular probe’s spiking activity. The second type is a hybrid dataset generated from the synthesis of real extracellular recorded data with manually superimposed spikes belonging to a number of single units \cite{rossant_spike_2016}. In the following we demonstrate that many of the t-sne generated low-dimensional clusters represent the activity of single units, while others group together spikes arising from a large number of putative units and likely noise. We develop a GUI that allows the fast visual identification of the single unit clusters and report on how accurately the manually selected clusters represent the labeled single unit’s in our test datasets. We then use the visual representations of spike clusters that t-sne generates to offer an overview of how the sorting/clustering problem’s difficulty increases with decreasing electrode density. Finally, utilizing the input agnostic nature of t-sne, we use it to embed the results of a new template-matching algorithm (kilosort) applied to the same ground-truth dataset. We subsequently use the GUI to overview and manually correct kilosort’s results and show that t-sne’s 2D embedding visualization makes digesting and curating the high dimensional output of automated spike clustering algorithms a simple procedure and also provides a satisfying “overview” of the otherwise overwhelmingly large, high-dimensional data sets. We will conclude with a discussion of possible extensions and future use cases of the t-sne algorithm for sorting and visualization of large-scale spike recordings.