Matt Sandy

One of the most interesting artifacts of #command-line based #Ai providers, is there is an incentive to produce a local script to process #data, as generating the code to fulfill a request likely is less impactful on their servers than processing the actual files to be processed.

Honestly I forgot now that it has been so long. Definitely a modified version of one of these, nlp.stanford.edu/projects/glo.... I plan on going through and building out my own model from common crawl data once some other pet projects are wrapped up.

Between #WASM, #DuckDB, #WebLLM, and some amazing progress on SQL specific LLMs, this could be an interesting era of lots of questions answered, and some answered correctly. Context truly is king.

The minimum temperature was all over the place this year, but some very high minimum temperatures.

Taking a look at minimum temperatures using #ggplot and #rstats.

DOGE.GOV Context Provider - Chrome Web Store Transforms dollar amounts on https://doge.gov/ into percentages of 20B.

New #ChromeExtension to give some context to the numbers posted on doge.gov. It shows the numbers as a percent of the $20B Elon took from the government.

fortune.com/2025/02/13/e...

Working on a Chrome extension to add some context to #DOGE whenever numbers get thrown around. Might also want to add corrections to descriptions of the programs.

Made some updates to little word guessing game. farmcow.app. Cosine similarity is now done locally to reduce latency, and the combinations don't include the answer. Pretty fun, pretty silly.

Just like people live in cities, pollution lives on fast thoroughfares.

Pulling out just 2024, looks way less cluttered and easier to see where we are in comparison to other years.

Taking a look at daily high temps in #Minneapolis. Using #rstats for this #datavis. Now to get the high, low, and factor in historical ice thickness at Lake the Isles to predict when the ice is safe to skate.

If you ever forget where the lakes and rivers are, you can just plot 311 reports until you gather up your shapefiles. #peopleliveonland

Winning votes normalized by land mass shows raised elevation based on the count of difference of votes.

Taking a look at the precinct level data using #rstats and #rayrender for #Minnesota. Elevation is determined by abs((Harris - Trump) / st_area) in each precinct and then rasterized, greyscale within boundaries is where the vote count was equal.

I always have time for fun R side projects.

Right now just trying with 2^4, seeing what has to go into the next box. It isn’t computationally efficient yet, but pretty fun proof of concept.

If you need to order moving boxes you can use #rstats to calculate how many based on best fit on a catalog of your items. #procrastinating

First post, first spelling error.

Alright, let's do this! Figured my first BlueSky post should be a banger, so I took a look at the setlist from one of my favorite bands, @themountaingoats.bsky.social, and created a visualization of the top 40 songs, and their co-occurrence within a setlist using #rstats for this #datavis.