Hacker Newsnew | past | comments | ask | show | jobs | submit | copperx's commentslogin

Are you using stepfun mostly because it's free, or is it better than other models at some things?

I think we are at this point where the hard ceiling of a strong model is pretty hard to delineate reliably (at least in coding, in research work it's clearer ofc) - and in a good sense, meaning with suitable task decomposition or a test harness or a good abstraction you can make the model do what you thought it could not. StepFun is a strong model and I really enjoyed studying and comparing it to others by coding pretty complex projects semi-autonomously (will do a write up on this soon tm).

Even purely pragmatically, StepFun covers 95% of my research+SWE coding needs, and for the remaining 5% I can access the large frontier models. I was surprised StepFun is even decent at planning and research, so it is possible to get by with it and nothing else (1), but ofc for minmaxing the best frontier model is still the best planner (although the latest deepseek is surprisingly good too).

Finally we are at a point where there is a clear separation of labor between frontier & strong+fast models, but tbh shoehorning StepFun into this "strong+fast" category feels limiting, I think it has greater potential.


I pay for copilot to access anthropic, google and openai models.

Claude code always give me rate limits. Claude through copilot is a bit slow, but copilot has constant network request issues or something, but at least I don't get rate limited as often.

At least local models always work, is faster (50+ tps with qwen3.5 35b a4b on a 4090) and most importantly never hit a rate limit.


> Claude code always give me rate limits

> 50+ tps with qwen3.5 35b a4b on a 4090

But qwen3.5 35b is worse than even Claude Haiku 4.5. You could switch your Claude Code to use Haiku and never hit rate limits. Also gets similar 50tps.


I haven't tried 4.5 haiku much, but i was not impressed with previous haiku versions.

My goto proprietary model in copilot for general tasks is gemini 3 flash which is priced the same as haiku.

The qwen model is in my experience close to gemini 3 flash, but gemini flash is still better.

Maybe it's somewhat related to what we're using them for. In my case I'm mostly using llms to code Lua. One case is a typed luajit language and the other is a 3d luajit framework written entirely in luajit.

I forgot exactly how many tps i get with qwen, but with glm 4.7 flash which is really good (to be local) gets me 120tps and a 120k context.

Don't get me wrong, proprietary models are superior, but local models are getting really good AND useful for a lot of real work.


I also started playing with 3.5 Flash and was impressed.

It’s 2× faster than its competitors. For tasks where “one-shotting” is unrealistic, a fast iteration loop makes a measurable difference in productivity.


Or Oh My Pi

How fast does it run on your M1?

Which things put you off?

He has some extreme takes on things, many of which I don't agree with, but I love that humans like him exist. He's one of the rare humans who has truly "sucked out all the marrow of life".


>He's one of the rare humans who has truly "sucked out all the marrow of life".

I think the biggest problem with him is that he styles himself as exactly that, but as the article points out he's a self-mythologizer who frequently makes stuff up. He's pretentious in the literal sense of that term, he's constantly exaggerated his own life, has even said that's the case and that doesn't seem as profound or counter-cultural to me as he thinks it is, given that we're mostly surrounded by exactly these kind of types now.


I take it as part of his act on film and in reality. He does it with a straight face and people are happily buying it and love him for it. Heck, I like him, too!

One can still enjoy the stories and not everything is scripted but that’s the fun puzzle to kind of figure out how much is staged and made up.


Lucky you. I always go "huh, so I wrote this?". And this was in the pre-AI era.

These feelings aren't mutually exclusive. I'm often like "I have no memory of this place" while my name stares at me from git blame, but that doesn't mean my intuition of how it's structured isn't highly likely to be right in such cases.

Like a painter not remembering a specific stroke, but being able to recreate it instantly because of years of muscle memory.

There is probably a bias here, because you notice the times where the code is unfamiliar more than the times when it’s still familiar. You wouldn’t go “huh” if not remembering was the normal case. If it were, you’d rather go “huh” if exceptionally you do remember.

It feels like that at first, especially as I get older. But I still think it comes back to me a lot quicker if I once understood it than if I was learning it from scratch. Possibly just because I know how I think.

God help us.

So GitHub Copilot will remove Anthropic as an LLM provider, I suppose?

I understand, but why does it need filesystem access in the first place?

so you can upload (er, provide access to) an image of a PCB you are looking to trace

Edit: I should have used a different word than upload. It's just old habit. According to TFA, there is no uploading. All processing is done in the browser, so the app needs local file system access to get at your image


There are other ways for webpages to get file uploads than this particular JS API. I upload files via firefox every single business day as part of my job.

This JS feature doesn't upload the file to a server. This particular app says right there on the page that it does the magic in the browser.

You appear to be misunderstanding on how browsers handle file uploads. You cannot get the local file path for a file. There is no C:\ or /Volumes or whatever your OS uses. Browsers deliberately mask that from the upload.


You can 'upload' a file into a completely local web app just fine. The directory access is only necessary if you need the web app to be able to spontaneously write back to the original file on your machine, or if you want to read a whole directory tree, which might be slightly convenient for things like gerbers but can easily be dealt with in other ways (especially with gerbers, which you can distinguish by filename, something that the web browser does expose to javascript).

(and I do think it's kind of irritating that Mozilla is fighting against such useful features on somewhat patronising 'the users won't understand what permission they're granting' grounds)


But I can drag and drop (or use the file picker) just fine on other websites.

To do what with? Upload? That's totally not the same thing as providing access for directly manipulating the file. That's basically HTML1.0 type stuff. JS file system access to provide a file to, I'm assuming, a WASM app is not even the same sport to be in the same ballpark.

Access for directly manipulating the file isn't necessary (perhaps if the files were enormous, but images and Gerber files aren't). One can upload/download files from a local web app just fine on firefox, and the WASM app can act on the file in memory with whatever APIs it wants.

> in memory

This, but it can also have the browser store it to disk by requesting the persistent mode from the storage API.

https://developer.mozilla.org/en-US/docs/Web/API/Storage_API


You are wrong. You can upload a file to a local web page for example, see this offline only image editing tool I made: https://lelandbatey.github.io/scrapbook_img_print_layout/

It does make it cheaper, obviously. But the barrier to entry is almost zero, like panhandling. That's why it can't substitute a job.

Stupid name? sure, but there's no point in fighting it. Claws is a sticky name.

These are all just transparent attempts to sound like "Claude", and if they're "sticky", that's the salient reason.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: