shunia_huang

It's more complicated in China not just because of people want a more safer environment, but also because of the culture and the social formation built by history.

Chinese were taught to grow in honesty and integrity explicitly for thousand years and it's becoming common consensus when more and more people are educated. When the bar raises and the gov is not given more freedom to the people, it's just nature to let the surveillance cameras grow to each and every inch of the land.

Me personally would chose safety over privacy though.

But they both spent tons of money on data collecting/labeling/generation, how is it bad compared to distillation? I thought their data are much better if they spent that much, and it seems they are stupid because with that much of resources putting in there with merely no output compared to the frontier models.

But Dario said (and maybe more ppl) that Chinese models are just distillation of their model and training data, so I guess your first point is invalid?

WOW the classic yes...but..., where's these statement coming from? I'm Chinese and I don't get the point here at all, you guys seem living in vacuum with biased news, and know much much better than we people living here, that's really confusing everytime I see/read comments lie this.

China now is open freely for almost everyone, come and see by yourslef, if you not living poverty.

The world itslef does not work the propaganda way whether you dislike or hate the gov for whatever reason. And yes you can criticize the gov as you like with the news you read otr the party you pro with, but never assume how the people feel and belives if you are not the people, and if you do then that's the same mindset as Trump or any other war-bringing president the US had, and you know what they did and still doing.

Or think it the other way, what if both A and O actually train with these data, then the enterprises found it(or maybe never), what's the other options? I'm not trusting these LLM companies because a. their model live with human generated data even if they claim to generate data with their own model, it's just nit human data, b. this is business not charity, eventually they live with customers' data, no exception

Sorry but it's not real BYOK to me as there is no UI/method provided to modify the providers, instead the app provides limited provider list and only allow custom API keys. I personally have a big list of providers under a CliProxyAPI service for easier management working with multiple coding agents and multiple serious or experimental providers, so I combine them whatever I want to have different experience, also much easier to work with so much API keys.

For example, I have my codex pointing to the CPA endpoint and the actual running model is a openai compatible one from an specific provider with specific base URL. I hope this could be considered add into roadmap.

I'm not reading the code on HN but I do second this. After implementing a enterprise level production app by myself months ago with vanilla typescript, I find it's much easier to think the vanilla way than adapting to React or things similar. Before this project I work with React for years and never looked the other way. The reason I chose vanilla typescript over React is for performance as the app requires tons of resources and I want to keep it minimal without the performance punishment, but ended with a much better experience for UI development.

You don't need to remember how to properly line up useState/useEffect/useCallback and all.

You could just insert any simple state object and be ok with it, or insert a well-known state management library and it still works the same way.

For UI libraries they almost always have a vanilla option, and even better, if you are not working with google sheet level of stuff, vanilla element management is much much easier to work with.

And no need to talk about the style frameworks, they always work the vanilla way.

With some default vite configuration, nothing you need to worry anymore and the framework level stuff are simple and easy to understand and is totally manageable.

While I'm trying to take a test it seems too much to start a Hello world application, or just to bundle my existing "dist" folder of a part time project.

Is it possible to install some tool, and give it a "dist" fold path and the application is generated all alone?

I'm stupidly lazy, and I hate work with graphical UIs to start a project and input that many fields, make me feel like visiting a 1990 web page requires me registering with tons of information and most of them are useless or at least I can provide later if needed.

fav-ed this post and will check again when another one poped on front page and says "local models is perfect now"

I would be punching my phone if the stupid network causing a wrong prompt and the LLM sends me unrelated answers. Correctness should be foundational no matter what, then improve the latency as best as possible. We all understand that if the network is bad then the latency can not be guaranteed but correctness should be.

I don't like how it reads differently based on perspectives, they should be crystal clear on what's allowed and what's not, but when they say '...unhappy about...' as a statement, it feels kinda amateur to me although I'm pretty sure the operators must be way capable than me lol.

It'd be nice to mention some big names here that are capable of:

a) git pull & push for updates

b) good enough CDN distribution, sometimes interactive examples/project page loads tons of files

c) good enough security promises of the entire platform/infra

d) good enough serviceable time, we do not need 99.9999SLA but better not down often

e) have generous free tier

f) great DX & UX, this one is small but small headache adds up quickly

Ah, wondering how long it will take Angular to replace it's sh*t building tool chain to fully vite compatible, hope it could happen before I change may career path or retire.

To me this is just "fake test". As I have said really world cases involves consistent IO loads and/or rendering loops, for example in my case I need to load tons of pixel data and decode them in works, then at the same time use canvas to render the decoded image and huge chunk of array data, they are real world high loads, there are tons of objects created during the process and way less counts than the "fake test", yet still optimizing the object counts made huge difference to the final performance.

Let's say talk about this in another more general case: virtual windowing. If anyone has tried to implement stuff and hit performance bottle neck and then find virtual windowing could help, it definitely involves two problems to solve, first is the UI responsiveness when more and more stuff got created and rendered, the object count usually should be way less than "10,000,000", yet still you could hit the wall.

I think I might be too negative about it, but I just want to share the real cases here.

I mean it's just not solving the problem right? If the model got injected then whatever the wrapper do can not prevent the consequences?

Sorry but am I missing something here?

As a non native speaker, I can even sense the little differences between these two.

I have answered something similar before, I struggle on sending messages as I want them to be received, with AI it is even harder, the "taste" of my thoughts, how I like to express, the habits of the phrasing or wording, get lost completely.

So I just never "AI" my content.

I've been long on JS but never heard things like this, could you please prove it by any means or at least give a valid proof to the _around 15%_ statement? Also by saying _quick enough to not be noticeable_, what's the situation you are referring too? I thought the GC overhead will stack until it eventually affects the UI responsiveness when handling continues IO or rendering loads, as recently I have done some perf stuff for such cases and optimizing count of objects did make things better and the console definitely showed some GC improvements, you make me nerve to go back and check again.

Blows my mind that you guys are so expensive lol.

I'm not so good at English but points are: - Websocket to send raw point data batch by batch - Strip the float value to integer if possible or multiple it before sending if it won't exceed Number.Max_Integer or something alike - The front-end should build wrapper around the received raw data for indexing so that no need to modify the data - There should be drawing/chart libraries handling the rendering quite well with proper data format with batched data

I believe the new UI libraries hit hard more than the AI impact. AI is not always that accurate so eventually if you want to deep dive in, you still have to turn around to the doc. But the new libraries though, they give the market another good choice, especially when shadcn came out, it's so huge that I personally even feels there's no need to go for the raw Tailwind experience, and what's worse is that shadcn is still evolving fast.

I believe the only way to let Tailwind survive is changing the business model.

Just want to add more context here:

For me personally, it's the uncertainty dealing with non native languages. I can see that for my own language it is so hard to take full control of all cases for even a simple word, that gives me nerve to think before I say anything non native, as I learn, there are more complex cases that makes a single word fitting to more and more scenarios. Although I totally understand that the native speakers will definitely understand even if I do it wrong, but still, I feel that nerve every time.

True madness for average user here is that they removed Intellisense plugins and force copilot but gives a seriously unusable quota for tab-completion, for non-paid users.

I just totally don't get it.

Yes and no, GitKraken actually have a graphical interface for WSL (or Linux generally), but it is barely usable as the WSL-g does not really work well. It's blurry for Hi-Res screen and the performance is like hell.

I would never try running any graphical stuff in WSL anymore, not worth it. VMWare with a graphical installation of any Linux system would be a preferred choice as I'm testing lately.

Graphical interface won't work well inside WSL, that's why I dropped my subscription on GitKraken and start using lazygit. lazygit simply works in almost any environment, and it works extremely well even if you are not into terminal stuff.

I tried this one: https://bobalearn.org/read/00000000-0000-0000-0000-000000000... And the words are mostly in Cantonese while few are Mandarin when listening to the sound, anything wrong here? After a page refresh it all goes to Cantonese now.

Also it seems that it can not read the whole story or did I missed the button somewhere?

I've not been able to communicate what I mean

As a native Chinese speaker that's always my confusion when communicate in English as I would feel that the word/phrasing can not express the meaning in my heart.

This is somehow real as I'm not so good at English but as a native Chinese speaker it feels a little bit hard to find the word/phrasing to express what I want to, but I guess your target audience should be more clear on what you are trying to express since it's the native thought process that the audience be familier with.

To me personally, the same meaning requires a lot of extra work to be expressed in English rather than in Chinese.

I'm in the same boat as a JS/TS developer for years to get used to npm install the recent AI command line tools under global scope, it's really weired. Personally I don't event install front end toolchains global as almost all of them work under project scope.

But I totally get the idea behind this, it's greatly combined with the cargo toolchain and is widely installed than cargo (which by the way npm is much easier to install than cargo with my personal experience), npm handles update nicely, and it can handle multiple arch too. I would still prefer npm install because I have installed a lot recently so it's ok.

So true when I'm running multiple agents in one project with multiple terminal windows. For example, with one working on implementing tests and another working on features, the feature agents will complain that the tests are not working and need fixing, while the test agent(s) will report outdated test coverage results due to newly introduced files.

It's annoying and hilarious at the same time.

Really user here: Switched from vscode to zed for ~2 weeks entirely in a windows PC, getting back to vscode today, just not feeling good enough: 1. zed updates frequently, but I'm not feeling any update at all. I mean, no new features, not fixing stuff that I'm experiencing that's not good enough; 2. while it put heavy in AI stuff, the ACP thing and the integration of codex and claude code just not working as expected as is. Especially I'm getting really poor outcomes from the same tool in zed compared to the cli itself, which is really frustrating; 3. the terminal in windows zed is barely usable, it's slow and slugish, sometimes the texts or some symbols are not rendering properly, it's just not stable at all. I really like to use the terminal inside the ide especially when I need to start up dev server, but with zed I'm used to open an extra terminal app when I need to start up dev server; 4. I'm a frontend dev, which means the tech stack should have great support in morden editors, but the case is not quite right in zed. I got jetbrain style inline type hints for typescript files with deferred types, but it is so annoying cause it makes a line too long I may even need to scroll horizontaolly more than may screen width to see the content at the end of the line, but the line has only 40 chars. At the same time, I'm not sure how to shutdown this feature after openning the settings. And when I want to set wrap width to 80 chars, I don't know how to do this either. The settings maybe there and are easy to tweak but it is not doing great to make the user understand it and use it; 5. it constantly showing some language servers are down and I don't know why, I do not like to tweak with settings so it could not be my bad, I don't event know how to touch the language servers;

I have more to input here but I forgot some of them, overall, I just want to get back to vscode which is much easier and battle tested. I think zed have a totally different perspectives on feature sets and stuff, that's good, so people could have another choice. Hope zed could do great and I will definitely come back one day.

PS: I don't think collab worth too much effort in an IDE, you have much better tooling out there and have better intergrations.