Perhaps a combination of the Matthew and the network effect. https://en.wikipedia.org/wiki/Matthew_effect https://en.wikipedia.org/wiki/Network_effect

Find good models here: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_...

Check hardware requirements here: https://rahulschand.github.io/gpu_poor/

Actually I am interested in the code. Did you use TikZ?

Nope, just plain matplotlib. The code was huge mess written under insane pressure, but I’m sure I kept it somewhere in my backups.

I also thought it was made with TikZ (which is a compliment). I used TikZ for my thesis: https://jeroenjanssens.com/phd/ and same thing here, the code is a mess.

It means Millions of tokens. Token in this context means either text token (as in tokenization https://en.wikipedia.org/wiki/Large_language_model#Probabili...) or video token where the paper describes them as "each frame in the video is tokenized with VQGAN into 256 tokens." (p. 6)

But why?

I'm a deep state employee of an as yet unacknowledged three letter agency whose propaganda blog received no traffic from this page until grandparent posted the link. I am a master strategist, playing chess in dimensions string theorists only dream of.

One of the better graph visualization tools!

Which tool do you people use most often for graph visualization? I kind of linger between networkx, Gephi, graphviz and d3.

I really struggle with this. I also use networkx, Gephi, and graphviz, and occasionally plotly. D3 is too much of a learning overhead for me at the moment.

I'm really in the market for a decent engine that's both easy to use and enables high quality interactive exploration of massive datasets. Plotly comes close, but struggles with larger datasets and is still a bit fiddly. I'd love to be able to lob a large structure into a UI and share with colleagues, and they'd be able to visualise and filter based on node/edge metadata, time series etc.

Something like Gephi, but web-based and considerably easier to use.

Agree. Neo4j’s visualization tools are surprisingly painful to use. I’d love to find something that allowed you to trivially add code to the layout engine. Graphviz’s clustering capability seems to be unique among any of the tools I’ve used as well.

Gephi can cluster, but the algorithm isn't that transparent or flexible. I've increasingly found myself doing the clustering in Python and rendering in Gephi.

That's why we built Graphistry :)

Can be used code-free (drop in a csv or log file) or via analysis workflows like from Jupyter notebooks and internal dashboards.

We are roughly parity for Gephi's features, and pushing (far) ahead on web, GPU, perf, ai, automation, etc. Always curious about gaps - our Slack is a good place to chat on that.

Heh, HN is a great place to fish. This looks great. I'll have a play around with the free account. Thanks for sharing.

I often use https://github.com/Rubjerg/Graphviz.NetWrapper to export the graphviz layout as json and then use d3 for rendering :)

I live in one of these houses and don't mind. Most of the time, you don't even think about it. In the few moments you do, it's just a reminder of what can go wrong.

Interesting perspective. Which academic field is that?

Difficult to say, I got my MA/Diploma in Landscape Architecture, then went into Environmental Planning & GIS, then Cartography & Data Science [1]. Yes, it probably is (still) a niche field and maybe it is only possible to do like I do because it is.

[1]: https://www.alexanderdunkel.com/cv_alexanderdunkel.pdf

People != Americans; orginal title of article is "Higher-income Americans are getting nervous about the economy".

A bit pedantic, but I agree the original title is distinctly clearer than the posted title.

Also, "getting nervous" is better terminology than "worried," since absolute consumer confidence is still higher among the higher-income group, it just decreased at a greater rate recently.