netghost

And here we are on HN discussing it.

I think what's most interesting is not that this works well for common languages, but it seems to be useful for languages that don't exist yet. If you want _some_ highlighting for your SQL variant, or your language that borrows many common idioms, or just sloppy code that's close enough to the target language, this could be useful.

You can play with it online. I was pretty impressed with the speed/quality given it's size.

It's definitely not going to replace a larger frontier model, but it's worth keeping an eye on.

I don't use them as coding agents, but they can be very useful for things like text transformation, summarizing, or text extraction.

That said, if you have a subscription to a paid model already, you're not necessarily winning out on anything except perhaps privacy, which isn't nothing.

TIL about `gum`. I love this and endeavor to integrate it into something sometime. Thank you for mentioning it!

They might have opinions about it, but look at the pay for frontend engineers at the same company. It's not uncommon to see the same seniority be 20% lower than a backend role.

I see a lot of people say you have to use methodology X, or that methodology Y is worthless.

I'm the end, I think we have maybe different uses for notes. Journaling, scratchpads, to-do lists, research, etc.

Take a methodology with a grain of salt. If it doesn't fit, there's a good chance it's solving someone else's problem, but you can always inform your own approach with it.

Looks great on mobile in portrait mode. Bonus points for the audio.

I think you're operating in a scale that is small enough that there's little risk.

You'll be able to iterate if you run into anything that doesn't work. You should however be clear on what problem you and your team are solving, and not just "get some rag".

That seems like a reasonable tradeoff, thanks.

Bun ships with lots of tools built in. It has support for bundling js, html, etc for the browser.

I suspect that if you want the best results or to hit all the edge cases you'd still want vite, but bun probably covers most needs.

Thank you. This looks like a nice improvement on pgtyped, and another good option.

I'm curious if there are any good patterns for dealing with dynamic query building or composing queries?

Out of curiosity, the post you linked mentions that it won't work for renames. What's the approach for these and other types of procedural migrations, such as data transformations (ie: splitting a column, changing a type, etc.)

With a declarative model, would you run the migration and follow immediately with a one off script?

Seems great for really small apps where you want your resource definitions colocated with the code using them. I'd imagine the benefits start to break down as your infrastructure gets more complicated.

The bigger answer is that if you're proficient and happy with CDK or anything else to wire resource up, you're probably not going to see much (if any) benefit.

Wonderfully done, thanks for sharing!

I think the make in the title is a bit misleading, the author is actually just advocating for having a consistent file you use for adhoc scripting and testing in your application.

The thrust of the article could be summarized as: if you type more than one command into the shell, make a script.

I think it really depends on how teams use their estimates. If you're locking in an estimate and have to stick with it for a week or a month, you're right, that's terrible.

If you don't strictly work on a Sprint schedule, then I think it's reasonable to have high variance estimates, then as soon as you learn more, you update the estimate.

I've seen lots of different teams do lots of different things. If they work for you and you're shipping with reliable results then that's excellent.

I did too, and I've had a challenging time convincing people outside of those ecosystems that this is possible, reasonable, we've been doing it for over a decade.

I worked on a product that was built around planning an estimation with ranged estimates (2-4h, 1-3d, etc)

2-12d conveys a very different story than 6-8d. Are the ranges precise? Nope, but they're useful in conveying uncertainty, which is something that gets dropped in any system that collapses estimates to a single point.

That said, people tend to just collapse ranges, so I guess we all lose in the end.

It natively supports vector embeddings, which seems like it could be nice. The sqlite extensions I've tried for vector embeddings have been a challenge to get working (may just be me though).

Hands down one of my favorite series. It's inventive, cynical, wry, dark, and entirely engrossing.

If you enjoy him, I was surprised by how much I enjoyed the Dogs of War series (1st and 3rd especially).

At some point you start to see his themes recycled across all these series, but it's still fun.

I'll bet you could take a relatively tiny model and get it to translate the transcribed "git force push" or "git push dash dash force" into "git push --force".

Likewise "cd home slash projects" into "cd ~/projects".

Maybe with some fine tuning, maybe without.

I found out that the guy who broke the thing I was working on was… me.

https://snakeshands.com - The best part of the snake…

Did you know that macos has had a per application menu that does this for years?

In most apps, press "cmd-shift ?" And you will activate the help search. I know, I know. We don't need help, but the other thing it does is reveal menu items from the native menubar.

It's great for finding fiddly commands in complex native apps.

I'll just mention that if you follow the link and read the article, linger a while.

Maggie Appleton's site is delightfully designed and thoughtful written. There are a ton of wonderful articles to enjoy there.

If this is comedy, sign me up for tragedy.

This feels like something that must be the root of innumerable small and easily overlooked bugs out there.

Maybe what's wrong with people implementing ai.

All I can say though is that I sure wouldn't want their bill after this gets shared on hacker News.

Kind of depends on your mac, but if it's a relatively recent apple silicon model… maybe, probably?

> Nemotron 3 Nano is a 3.2B active (3.6B with embeddings) 31.6B total parameter model.

So I don't know the exact math once you have a MoE, but 3.2b will run on most anything, 31.6b and you're looking at needing a pretty large amount of ram.