Noah Veltman
9.4K posts

Noah Veltman
@veltman
Data+viz @netflix. I try to make fun things on the internet. Code, graphics, data, journalism, etc. Previously: @datanews @bbcnewsgraphics @[email protected]
California Katılım Kasım 2008
564 Takip Edilen6.8K Takipçiler

@wallacetim makes me nostalgic for blocks.roadtolarissa.com/veltman/4d1413…
English

@simonw is it just images that are an issue here? seems like you could have it render a regular MD link where the link text is "Read the docs on MDN" or something, but the link goes somewhere else and exfiltrates in the process...
English

Once again: if your LLM chat system has access to untrusted content (in this case GitHub Copilot Chat can see the content of text files in your editor) do NOT include support for rendering markdown images from arbitrary domains, or you'll have a data exfiltration security hole
Johann Rehberger@wunderwuzzi23
Another data exfil vulnerability in an LLM app fixed! 👨🔧 This time in GitHub Copilot VS Code Extension. #infosec #llm embracethered.com/blog/posts/202…
English

@dmccarthy7 all of the Big Three and the Big Four are pretty high on the list!
English

@veltman McKinsey may not have the same raw number, but their slugging percentage is up there

English

@adamrpearce @tophtucker dang now I gotta read the CA high speed rail biennial reports to make a three-axis version (dollars, years, miles remaining)
English

@tophtucker what about actual time v. predicted time remaining? (might not work as well w/ more jumps)

English

@hamiltonulmer "the perf issue arises when there are a lot of these ... and even then it's just loading time" - is that because you're generating a new interpolate() for each? or is it slow upfront even with a shared interpolator?
English

@veltman will email you
the perf issue arises when there are a lot of these, and I'm using a small max segment size. And even then it's just loading time where it's obvious, not the actual animation
English

@hamiltonulmer and because of the radial symmetry you can kind of custom interpolate by angle + radius instead - this is what flubber toCircle does (alas there is no toEllipse)
English

@hamiltonulmer I suspect that it's all the upfront path.getPointAtLength calls that drag, since your shapes are all nice and symmetrical you could precompute the corresponding polylines in pure math (and it would even look a wee bit nicer to a discerning eye)
English

@hamiltonulmer other hunch is precomputing the frames since there aren't too many, and then instead of interpolating you're only swapping (and then maybe you could even do it with a spritesheet + changing the offset!)
English

@hamiltonulmer for the ellipse, I might l just swap the real arc path in whenever t was close enough to 1 (or zero), since I don't think that would cost any more
English

@b0rk almost ten years old but I once wrote a "Command Line Murder Mystery" as a teaching tool github.com/veltman/clmyst…
English

what helps people get comfortable on the command line? jvns.ca/blog/2023/08/0…
Would love more stories of things that helped you in the last ~5 years!
English

@veltman lol I know you're joking but
- sveltekit for app framework
- d3 for data utils
- duckdb for data plane
- trpc for typesafe APIs
- tanstack for virtualization & server state
is a kind of an intriguing stack for a data-intensive application
English

Fantastic work by the team on this release. @sperosck in particular integrated a new custom date range picker that works much better than the native browser data picker.
I think this work underscores the advantages and pitfalls of building mature applications with @sveltejs ...
Rill Data@RillData
The .28 release has refinements to the dashboard's concept of time to help you understand how measures and segments are changing. ✅Ability to hide measures & dimensions ✅Custom date range picker ✅Support comparison ranges for ‘all-time’ Learn more: docs.rilldata.com/notes/0.28
English















