mooreed
Joined 14 years ago
Joined 14 years ago
However, after a VERY short perusal, I grew a giant sense of empathy for non-native English speakers. The readme is gentle enough to English speakers (aka: +95% English) no less I felt like I muddled through renaming tokens in my mind as I went. However, that quickly showed me two things.
1. It reminds me why I never seem to finish classic Russian literature… so often get lost in the introductory parade of names that are a cache miss for my usual set of names.
2. This is perhaps a significant cultural muscle that has never been necessitated for English speakers. Since the earth has largely been using English (in some capacity) for significantly longer than my life span - as my favorite joke says in the punch line “what do you call someone who only knows one language… uni-lingual… jk: American”
PS: it seems like there could be an open registry maintained by “Americans like me” who would rather pre-process the code for tokens within the docs and src… seems like a “DefinitelyTyped style” definitions registry would be very niche, but SUPER useful.
I also would love to hear more about the cluster shapes and cardinality of the coordinate system. I consider myself am pretty versed in data analysis, however with less expertise on NLP topics (eg t-SNE).
So a quick blurb like: the units on the axes in the graph are “a reduced embedding space” designed to keep structure and to reduce the dimensionality such that the clusters could be plotted on screen…
(I’m not even sure that’s correct, but I would have loved for you to have informed me on the one sentence visualization choice and then point me to t-SNE.)
Overall nice project - and it reminds me of a painful professional analysis lesson I have had to re-learn more than once.
> After working for NN hours on an analysis, and finally breaking through and completing it, overlooking the title and labels is the biggest footgun I have ever dealt with.
[1]: https://github.com/segmentio/ksuid which has very similar use cases.
Or maybe even a way to discover related statically typed definitions based on the validation rules?
It would be really nice to not define parts of a data model that provide little to no business value - but where you can easily “stub your toe”.
> “you can solve every problem by adding or removing an abstraction layer”
Have you considered generating your YAML/JSON config with something that composes?
If you are open to it you might be interested in dhall [1] as it’s a config language with variables, functions and imports.
I have used it for pet projects and I could see how it could offer some tidy encapsulation patterns for larger, more complicated production applications.
So far the human signal was that text and images were crafted by humans, of course with some tooling.
However as the universe of images and text corpuses grow, they will embody the idiosyncratic nature of the AI generation process. And those “glitches” as has been called higher in the thread, will potentially get fed-back into the training data set with out some filter using a human-signal, if even just an AI human-signal.
See https://nopecha.com/ (I’m not affiliated) and plenty others.
(Added edit) I too agree with the sentiments that a v2 should have a finer grain resolution where the data model would recognize a bill as a collection of clauses or statements. And attribution and intent of those clauses would be transformative (I think) for regular citizens.
Because then the internet would be able to track unintended consequences by author. This is arguably what the internet does best.
Have no fear Gen Z is a pendulum towards radical 90s colors.
https://link.springer.com/content/pdf/10.1007/s00371-021-022...
[ my public key: https://keybase.io/mooreed; my proof: https://keybase.io/mooreed/sigs/-i_S91JTQoy0oMr_WirgBLVZbD3IkaN2cPV8HjcqUb0 ]