Robert Lesser

What are some great datasets hosted on Hugging Face? We just added a way to quickly import, embed, and build interactive visualizations from them. Would love to get some hidden gems in. E.g. here's 650,000 1933 newspaper articles from LOC, extracted by Melissa Dell. atlas.nomic.ai/data/nomic/a...

I think that’s true, and also people find ways to rationalize support for things they genuinely enjoy using/doing

I think trending is still up? observablehq.com/trending

But it’s been massively hidden, I’m not sure if there’s a single public-facing link to it on the whole site. Definitely a bummer

For Maurice DiMaggio, an electrical contractor, morning drives from Matawan, N.J., into Manhattan via the Lincoln Tunnel usually take two hours. Today's took only one. "I would rather have an hour not in the vehicle, in traffic," DiMaggio, 45, said from inside his work van. If congestion pricing proves a permanent "commute clipper" for DiMaggio (who is related via his grandfather to Joltin' Joe), he is all for it. "I can't run my business off the subway, but maybe this whole congestion pricing thing is keeping people from driving."

every single time, congestion pricing becomes way more popular after it’s implemented

traffic sucks, but people refuse to believe it will go away until it actually does

To absolute’s credit, nowhere else I know of let you get a Thai iced tea along with your bagel. That was really something

This is a real loss, but my undisputed UWS GOAT Broadway Bagel on 101 remains very much alive. Long live the egg everything BEC

My sense is that both absolute and relative time could be meaningful over time - “what does decay look like for 10 year time gaps” but also “what does the decay trajectory look like for sites from 2014?”

With enough data you could make a nice “Medusa chart” showing both. Cc @toph.me

You know a side project is getting out of hand when you start making a settings page before anyone has actually used it

A heatmap of years and states, showing a large blue cluster in the center where Southern states vote Democratic. and other geographical patterns.

A heatmap of years and states in alphabetical order, showing no clear global structure

New blog post! Updated for 2024, my favorite example of why alphabetical ordering is bad for geographic features -- US presidential results since 1828. The left image shows regional patterns in a geographic ordering that the right (alphabetical) simply loses. benschmidt.org/post/2024-11...

What a beautiful win. The look on Ryan's face in that final moment made the whole season worth it!!!

Adventures in DuckDB + Huggingface + ArrowJS - If you try to stream arrow IPC out of HF with DuckDB, for some reason the batches come back in ~random order! ArrowJS decided to explode when this happens.

Not even sure who's at fault here, but excited for this ecosystem to continue to mature #databs

Clippy, the godfather of generative AI

This problem was solved years ago

TMD Web Team Alum Starter Pack 👀

I am a broken record on this but LLM text embeddings are an incredible breakthrough, and the ability for anyone to build pretty good classifiers with structured output could be insanely useful.

Trying to build NLP interfaces is taking my team an extremely long time and is extremely brittle

Back in 2022 I published this post analyzing 15M tweets with Wordle results, with some fascinating results: observablehq.com/@rlesser/wor...

The hardest part by far was gathering a huge dataset from a platform hostile to such analysis. Very excited that Bluesky encourages this type of exploration!

Probably the getRelationships endpoint not working correctly: github.com/bluesky-soci...

Great write up on how Val Town built Townie, probably the best LLM coding experience I’ve used.

A huge part of what makes it so nice is how little boilerplate/infrastructure the VT environment needs. Easier for people and easier for LLMs when everyone can focus on the business logic alone.

One of the most exciting things I’ve worked on at Nomic.

There is huge untapped potential in taking people on a journey through a dataset, especially for text and image sets that are currently so hard to reason about

This is now a force-directed-graph-posting account, all other content posted is anomalous behavior