I’ve built a small suite of honey-do automations for public defense attorneys in Washington (selenium scripts and OS calls to do docket intake / organization)
Currently packaging as a windows desktop app, recording a demo video, and then distributing
Very cool. I jumped in here thinking it was gonna be something else though: a packaged service for distributing on-prem model running across multiple GPUs.
I'm basically imagining a vast.ai type deployment of an on-prem GPT; assuming that most infra is consumer GPUs on consumer devices, the idea of running the "company cluster" as combined compute of the company's machines
Commenting because I can't get everyone's email; if you're reading this thread and would be interested in an automation engineering conference please email me at mailto:[email protected]
In my experience chatGPT solves the challenge handily, but it does get to the level of problem that GPT requires my code review for, so either I'm not as good at using chat as I could be or they've found the point at which someone has to prove they can actually do code review to filter; nice.
I don't see how I (or LinkedIn) is making you do anything? LinkedIn is a place I can post data. I choose to do so in an attempt to market my resume. I fully expect that the data I post on LinkedIn's server becomes and is the property of LinkedIn, and wish it was more effective at extracting value from it?
Because LinkedIn is less effective than I'd like, I support 3rd parties scraping the data I posted there, again on the hope that they'd be more successful at marketing that data, which I would benefit from as the data is my resume.
Okay, so 30 million people, or more conservatively 16 million people. Same question, and before you say anything, don't be condescending.
ETA: Maybe most conservatively, let's use only the % uniquely included in U-6 and excluded in standard, or 14.4 million people. I'll claim these 14m people are the "gatekept from full employment" in that they don't qualify as narrowly unemployed unless you include "all people marginally attached to the labor force, plus total employed part time for economic reasons"(1)
Same question. 14m people who are being excluded from labor, are they not free to attempt to generate income via means other than labor, lest they suffer the judgement of Abraham?
In other words, is Abraham hiring? If not, are the people he refuses to employ meant to accept serfdom to preserve their soul?
OK, so what is to be done when access to labor is gatekept? If you can't have a job, or can't benefit from a job beyond "give a man a fish, he eats for a day", are you not meant to look for an opportunity to generate income outside of working?
I sure do! If LinkedIn can't market my resume to open roles then letting recruiters roll their own scrapers against it is the next best thing. I understand that LI owns my data, I just wish they were effective in using it!
A webscraping / data pipeline to get the .pdf "Explanation of Benefits", "Proof of Coverage", and "Drug Formulary" for every Medicare Advantage plan in the USA
These docs are gonna be used in a product for medicare brokers (if you are/know one please reach out open enrollment starts Nov.1!), and the pipeline is horizontally scalable to ingest updated 2026 plans overnight @ start of open enrollment (though some companies are posting updated plans earlier)
There are some clever tricks at play but mostly it's bog-standard browser automation; I'm also in an interview process with 2 entities (one funded startup and one massive corporation) talking about web automation roles, and while it's frustrating that they're moving so slowly it's working out to give me time to build this well.
I see your point for technical users, what I'm imagining is something where I install NixOS, and then download LibreOffice via a GUI, as is done on Windows/Mac/Most Linux tbh
I don't think an LLM is required here; for a very significant subset of nixpkgs I think updating the configuration.nix is a deterministic process.
I’m reading your summary versus the other article here, but it seems like for writing code, Claude would be the clear winner?
When chat breaks apart for me, it’s almost always because the context window has been overflown and it is no longer remembering some important feature implemented earlier in the chat; it seems based on your description that Claude is optimizing to not do that.
Thanks for this; I'm still not getting it, I think because I haven't really deployed software at scale before? It's a CI/CD solution?
I can see how this would be instrumental for coordinating hundreds of developers across tens of products, but as someone who's only ever worked as tens of developers on one product it seems like a lot of overhead for really no ROI; what's the difference between using this product and having a DevOps team that coordinates CI/CD in-house? Is Palantir's edge here being the out-of-house devops team for any large enterprise?
Do you think a "batteries-included" NixOS install could be developed for users like me? I feel like a NixOS with libreoffice, zoom, etc. could be a drop-in windows11 replacement for a lot of users.
Maybe the key feature would be a graphical installer; something that presents an app-store iterative-install UI and in the background handles configuration.nix, so nontechnical users don't have to use the terminal or wrap their head around declarative config / version control to get a reproducible desktop that they can simply install on new machines.
Reading this comment I can't help but imagine a high school student using the same pattern to respond to an "open period" being changed to "study hall" with mandatory in-library presence; which is not to dig on you, just to raise the idea that maybe k-12 education really is a conspiracy to train people to sit in factories.
The first feature I wanted to add was a "Quote from xometry" panel; I cloned/GPT'd for a couple minutes and found that actually adding this would mean pulling tricks with selenium that don't scale.
Have you reached out to xometry/hubs.com/other Print-aaS companies for a partnership?
"set up a cloud instance ahead of time and just boot it up"
That sentence is the closest to what I'm imagining--the thin client to personal computer, your use case is interesting (hackathon peripherals); I'm thinking of this as a lifetime-durability play.
Basically what if my home's NAS could also serve all my applications, and at any given time I really just have a KVM into it. If we assume network is fast enough, that would achieve the UX I want, which is that the KVM is effectively disposable and the personal computer stuff all stays intact.
Currently packaging as a windows desktop app, recording a demo video, and then distributing