The uv tool is so small and fast that this all happens in under a few seconds!

The download script git-pulls the repository like normal. Next, it installs the uv package manager and immediately runs it to install python dependencies in a local virtual environment, based on requirements listed in the repo's pyproject.toml file. Finally it installs and starts Jupyter

This template helps you set up a download script hosted on GitHub. A special terminal command can be used to grab it from github pages, and immediately run it. It looks something like this:

curl -fsSL path-to-the-download-script.sh | bash

The Sony ZV-E1 has animal eye autofocus. Same sensor as FX3 and A7SIII. So, good for video but low res for pics. Of those, only the a7SIII has a viewfinder

I used Figma, but just because I know it well. I think excalidraw or figjam would be great too

Hmm yes, I should learn more about UML. Though I can already tell its primarily focused on grouping stuff into boxes with lines between them

The important thing for me is understanding lifecycles and data flow. If it was just boxes representing classes or files, then I wouldn't find it nearly as useful

I still firmly believe LLMs aren’t conscious and there’s no clear path toward artificial consciousness. But perfect simulation of consciousness seems…possible

The perspective that “this is all still curve-fitting” can still be applied to modern AI. I think.

But that begs the question: what happens when an artificial intelligence applies curve fitting to every facet of human existence?

And so, I'm questioning if "science communicator" is the term to use. Science itself needs to be communicated in mostly niche circumstances (climate chance, conservation, etc). Otherwise, science will influence your life if and when it damn we'll feels like it

Her: What do you mean?

Me: It's about misunderstood incentives. I don't think society is healthier if more people know how black holes work. I do think society is healthier if more people are intrinsically interested in how black holes work.

I was suprised to see gpt 4 quote data from a paper as "between 1011 and 1014 photons per second". You'd think it would have realized the misrepresentation of scientific notation: 10^11 and 10^14. If it can't pick up on this, then what else?

Yes! I hope to explore this for the future of a academic publishing. It should all just be a graph-like datastructure, with figures at different levels of abstraction. Ready to be expressed through whatever ai-generated modality you wish to interact with (natural language, multimedia...)

I imagine a cultural line will develop between people deemed to be producing 'fake art' and people producing 'real art' (with generous helping of AI assistance). 'Fake art' peddlers become scapegoats.

Hard to imagine because the line becomes so murky

The book 'wanting' by Luke Burgis has a nice history and analysis of the scapegoat mechanism. Plenty of examples throughout history of people in power choosing and architecting a scapegoat

It's all micropipettes now, right? Show me an emogi for this bad boy

Thank you!