This is what opencode does for me. One harness for all models, standardized TUI, and they're rolling out a product to serve models via API with one bill through them
What are the new file format initiatives you're referencing here?
This solution seems clever overall, and finding a way to bolt on features of the latest-and-greatest new hotness without breaking backwards compatibility is a testament to the DataFusion team. Supporting legacy systems is crucial work, even if things need a ground-up rewrite periodically.
There are also a bunch of ideas coming out of academia, but I don't know how many of them have a sustained effort behind them and not just a couple of papers
I’ve been excited about lancedb and its ability to support vector indexes and efficient row level lookups. I wonder if this approach would work for their design goals and still allow broader backwards compatibility with the parquet ecosystem. Have been intrigued by Ducklake, and they’ve leaned into parquet. Perhaps this approach will allow more flexible indexing approaches with support for the broader parquet ecosystem which is significant.
Lazy trees is a long awaited change for many, without it flakes are essentially unusable in monorepos. As a one-time Nix user, Flakes definitely don't seem like an ideal solution but the Nix community lets perfect be the enemy of better too often, and it has the largest following of all the solutions out there (niv being another). Given that, I hope lazy-trees and other improvements that make it actually usable get merged into Nix upstream.
If the NixOS/Nix maintainers don't like flakes/DetSys (which I think is somewhat valid!), they need to put forward a canonical way of pinning channels, otherwise there's not really an alternative standard for the community to build around. The usual answer of just use niv or some other esoteric solution that a subset of hardcore Nix users like isn't enough.
I’m building a (small) monorepo with Niv. It is much easier to understand and use than flakes. Flakes are simpler to use directly for sure (eg from CLI), but if you actually want to use them as dependencies, it becomes way harder to understand.
All Niv does is produce a JSON file with your dependencies, then you use the Nix functions they provide to read and load the dependencies. It is probably the simplest possible dependency management strategy in Nix.
This is also not a Nix specific strategy tbh. If you are using Python, Go, JS, etc they manage dependencies in the same style (lock file with dependencies, tool to update them).
I don’t think the documentation for it is any worse than the rest of Nix (low bar, but we are comparing things in the Nix ecosystem).
Flakes seem a lot more magic under the hood, whereas Niv is just providing the arguments to existing Nix functions. They need special handling for nested dependencies (“follows” is a bit weird and hard to discover). With Niv, everything is just a function evaluation. Dependencies source caching is non-obvious for nested dependencies, with Niv it is more explicit (although you do need to manually instantiate them, but I prefer the explicitness).
Nothing major, but it just adds a bit of cognitive load.
Naomi Wu has been talking about this for a while [0]. She lives in China, and has flagged this multiple times on Twitter, apparently without much response from the Signal team to her frustration. The issue seems to be that Signal simply sets the incognito flag when users are typing in the app, which kindly requests keyboard apps to not track the input. Unfortunately, since most American made keyboards for chinese writing are slow and difficult to use, most Chinese citizens use keyboards made by companies made in China, which does not provide much reassurance w.r.t adherence to the incognito request. Your concern seems to be spot on, and it'd be worth Signal at least making this clear to users when they first begin using the app.
While the answer from most technical users will be "just switch keyboards", it's also widely accepted that defaults matter and unless Signal makes an active effort to discourage people's default choice, the larger purpose of privacy will be somewhat defeated.
> You want to write Chinese, you need an IME. Apple/Microsoft/Linux/Google all have their own IME, but they aren't very good- typing on them is SLOW. Most Chinese use [...]
So, the ask is that Signal (a western company) develop a Chinese keyboard that is better than the vast majority of Chinese keyboards on the market (something Wu says is hard for western companies), and then force all Chinese users to use it, even if they prefer something else? What happens if they develop a bad one?
Keyboards aren't that special here. Signal opens up links in web browsers and some users install insecure web browser (also more commonly in China than in the US). Signal runs on an operating system and some users install insecure operating systems (also more commonly in China than in the US). There's no logical stopping point here until the whole device is Signal.
I think it's good for Signal to try to do one thing (encrypted communication) well, rather than trying to do everything.
While assurances from the founders make (classically) liberal idealists all warm and fuzzy inside, they aren't worth much until there is a controversy to use this post as a measuring stick on. At that point, we'll have a more clear idea of exactly where they draw the line, which I'd be interested in knowing as a supporter of the company. Regardless, Substack seems to have high quality built in off ramps to the point where deplatforming someone from Substack won't have the same memory holing effect that it does with social media.
The best (and worst depending on which side of this future controversy I find myself on) part of Substack is the lack of switching costs. No one is staying on Substack for their barebones content editor, and retaining your own mailing list (and now custom domain!) makes firing up a Ghost newsletter easy. Their other value adds like legal support and revenue advances will most likely never be given to anyone of guilty of sufficient offense to be kicked off the platform, so that point is moot when it comes to the limits of their free speech ideals.
Optimizing for efficient, capable teams is a difficult problem, and seemingly untenable above a certain size by nature. With that in mind, the obvious answer (especially in software) seems to be to keep teams as small as possible. Netflix's strategy is higher half the people, pay them twice as much, and give them as much autonomy as possible according to the CEO's new book. The obvious counterpoint to that strategy is it leads to overwork and will inevitably devolve above a certain size as well.
For a more radical strategy on keeping company size small, I like how the startup Stedi has taken a crack at innovating on this. Their entire tech stack is structured around AWS offerings, mainly serverless. The idea is that by paying a premium to Amazon, they can outsource all server management, scaling, etc. issues which typically take up tons of engineering bandwidth. Presumably, Amazon's round the clock experienced team will also be able to handle it more effectively than they could. In the meantime, Stedi can now avoid hiring for tons of jobs and keep headcount as small as possible, enabling engineers to only work on things that matter.
While I'm not sure how Stedi's strategy will play out, it is one of the few opinionated stances on company organizational philosophy that attempts to take advantage of recent evolution in cloud offerings, and internalize the consequences of those improvements.
https://www.nutritiousmovement.com/furniture-free-ahs13/ is an example of someone doing this in their home. Personally, my setup is a coffee table along with a zafu (buckwheat hull filled floor cushion) and sheepskins or a zabuton to cushion the floor a little. It's very comfortable, although I'd like to also have a standing desk to go with it. Coffee tables tend to be a pretty good height for this purpose if you want to avoid purchasing something custom. Besides that it's just a regular desk setup, albeit missing drawers or storage on the table.
Yeah, here is what I'm working with as of now. Moved recently (like everyone else) and still getting an office setup, but this is the basics. As far as height, I'd say it's a very ergonomically sound setup when in kneeling position with the cushion in between your feet. Laptop just below eye level, arms level, etc.
I think the biggest benefits are it forces you to move and adjust a little bit more than in a chair, and forces you to use your muscles to sit properly much more. Would highly recommend to anyone.
https://ibb.co/XYN3HFx
I was browsing though the different pillow/cushion and was a bit worried at first by the $150 zafu/zabuton but it looks like there are ~$50 ones so I can give it a try.
Scott's book is about how bottom-up complex systems are destroyed and reconstructed as top-down centrally planned organized systems in order to become more "legible" for the destroyer (the state). The general theme is the state can't understand how certain traits of the system are actually adaptive because it views from the outside. These traits are seen as irrational, so the state destroys the system and rebuilds it in order to make it rational, and thus understandable and legible in order to optimize for extracting a resource of some sort. Legibility in this context is used almost pejoratively to say most beneficial elements have been destroyed to optimize for one parameter, often ineffectively in the long term. Scott Alexander has a pretty good review if you want a deeper dive into the book's examples of this process.
https://slatestarcodex.com/2017/03/16/book-review-seeing-lik...
I think you would be very interested in Roam Research if you’re ever looking for a high tech solution again. Only recently entered public beta, but it seems to by nature organize notes very easily and fluidly.
https://roamresearch.com/
