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?
It is a CD solution, there's no CI involved at all. However, it extends the 'delivery' portion of that phrase to also include so called "day 2" Kubernetes-centric functionality such as "how does my Secret content get populated?" and "how to I apply only certain values.yaml to certain Helm release versions?" Then, Mission Manager also gets into the game of "how do I ensure my cloud resources are present alongside the software?"
It also gets into the compliance side of things (arguably it's all about compliance) in that new releases, config changes, vulnerability mitigation, downtime windows, and a ton of other stuff are optionally gated by a review process. If you want to change the values.yaml for an existing release, you can choose whether that is a 4-eyes change
That latter bit also speaks to your question about "who is the target audience for this?" because, like many things Palantir, it's a lot of the defense industry and intelligence community, as they care a lot about having gates and controls in place. Apollo specifically also gets into "edge deployments," such as what software is running on a laptop that goes out into the field, comes back to base, and only then gets its updates pushed to it
I detest them with all my heart, but I'll be straight that Apollo does actually model the deployment concerns of software very well. Every place that I've been always hand-assembles their own CD story using off the shelf stuff, and so if one has compliance needs then Apollo allows to pay to have the "hand assembled" problem go away. That said, I have _absolutely no idea_ how much Apollo costs, so I'd also speculate that most organizations would be much happier just paying for a competent DevOps role to do the best they can to use off the shelf tools to hand assemble their own CD story, or use BigBang[1] which is at least open source and a standard (one may not like it, but it is at least consistent)
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.
For stuff that's not in Nixpkgs like this, IMO flakes is the best option. Both of these require flakes.
On the topic of deploying flakes to consumer users. I'm currently exploring using https://github.com/nix-community/nixos-generators to create pre-made install iso's for a full "batteries included" experience. You can just "overwrite" the regular nixos image. Its Nix all the way down.
I was surprised (and impressed and excited) to hear some rumblings about some of y'all thoughts "provisioning" at NixCon. I still am getting my head around clan, but I have long admired your work and wish you a lot of luck.
Def post the link when up; from your website what I'm getting is somehow docker images + kitchen sink, which may just be my lack of knowledge but in any case means I as a consumer / target user don't get it
You're misunderstanding, and it's a great opportunity for me to clarify:
I want a graphical installer for applications within NixOS.
Currently NixOS applications are added by editing the configuration.nix, with more specific tooling avaliable via flakes and home-manager. We agree that this is cool and good.
What I want for consumer use is NixOS, with an interal graphical installer that handles updating configurations in the background; ideally forming a no-terminal-necessary UX for consumer users.
Ah, yes, I see now. I think a NixOS configuration is best managed as text, but I can see now how code generation could potentially bridge that gap. I'm not currently confident in the ability of LLMs to do this, but I do think that a kind of "NixOS IDE", perhaps with a some GUI elements and writing to some constrained files is a good idea.
Once upon a time, Snowflake OS was working on this, but I don't think the project is very active anymore.
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.
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?
That’s part of but not fully it, thanks for the validation that it’s not just me re: onedrive.
I think partially I’m wrestling with the almost philosophical question of whether Linux is viable for consumers, because ideally I’d be a consumer, but I’m so frustrated by switching costs between windows/mac that surely open source has to be the solution, and loop from this point.
There are, in my mind, 3 kinds of data loss when my laptop breaks:
1) passwords
2) files
3) applications
(1) and (2) are at least in theory solved by simply embracing vendor lock in. In practice you really can just keep buying Apple forever, or figure out OneDrive for real and it will work.
(3) is where I think I’m breaking with the current state of the industry, but it is not mandatory imo that I wouldn’t be able to get my vscode, pgadmin, cad application, office suite, etc. all downloaded with their config intact on a new machine.
One answer to some of those problems is to use remote desktop technology.
Last time I went to a hackathon I brought a 15 inch Alienware from 2017 which has bad connections in the USB system and is on the edge of death. I loaded up Visual Studio and the Unity Framework ahead of time so I'd be ready to use the same tools as my team.
Personally my favorite hackathon kit is a tablet plus a keyboard and a mouse. Remote desktop into a big computer and you have the sleekest kit anyone has and the most powerful computer. I have a powerful computer at home but I have ADSL, it is possible to remote into but latency is pretty bad.
My plan, the next time I go to something like that, is to set up a cloud instance ahead of time and just boot it up. Somewhere between $1-$2 an hour would buy a powerful machine which would really be a bargain if I only want to run it for 20 hours on an occasional weekend.
"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.
Remote: Yes, prefer in-office though
Willing to relocate: No
Technologies: Fullstack react, express, node, postgres; data pipelines python, kafka; testing selenium
Résumé/CV: http://www.linkedin.com/in/douglasemckinley/
Email: [email protected]
Heavy preference for in-office roles in Seattle. In my experience the 10-minute call is still king for mutual evaluation, so here's my Calendly: https://calendly.com/dougla/conversation
Don’t you already want to hit the baseball as hard as you can at every opportunity? Just with the caveat that you need to develop skill with using a one-size-fits-all bat? Is bunting and going for balls that big of a deal currently, that a player who could rock one into left field would decide not to?
Surely any strategy around loading up bases to stack the deck for your strongest hitter remains, it seems like this levels between hitters more than from hitter to pitcher?
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.