Scramble: Open-Source Alternative to Grammarly(github.com) |
Scramble: Open-Source Alternative to Grammarly(github.com) |
>If you have any questions about this privacy policy, please contact us at [your contact information].
easy to install in LibreOffice
Come on.
Pitch this honestly. It'll save me clicks if I'm using an LLM to checker grammar already, but if I use Grammarly it's not an alternative at all. Not by a long way.
Edit: The author replied to another comment that there is an intent to add local AI. If that is the plan, then fix the wording until it can actually be considered privacy-respecting: https://news.ycombinator.com/item?id=41579144
To the best of my knowledge all localhost connections are exempt from CORS and that's in fact how the 1Password extension communicates with the desktop app. I'd bet Bitwarden and KeePassXC behave similarly
Edit the 'background.js' file in the extension and replace the openAI endpoint with
'http://your.local.ip.addr:5001/v1/chat/completions'
Set anything you want as an API key. Now you have a truly local version.
* https://github.com/LostRuins/koboldcpp/releases
* https://huggingface.co/bartowski/Meta-Llama-3.1-8B-Instruct-...
* https://github.com/LostRuins/koboldcpp/blob/concedo/kcpp_ada...
https://help.openai.com/en/articles/7730893-data-controls-fa...
So, it's not as clear cut. The general approach of using LLMs for this is not a bad one; LLMs are pretty good at this stuff.
Key features: - Uses your OpenAI API key (100% local) - Pre-defined prompts for various improvements - Highlight text and wait for suggestions - Currently fixed to GPT-4-turbo
Future plans: add LLM provider/model choice, custom prompts, bug fixes, and improve default prompts.
It's probably buggy, but I'll keep improving it. Feedback welcome.
I like this approach so much better than leaning on AI because it’s more my “voice”.
Grammarly's core functionality is not even LLM-based; it's older than that. Recently, they've crammed in some LLM features that I don't care a snoot about compared to its core functionality.
This tool, like any other "Grammarly alternative," is just another GPT wrapper to rewrite my text in an overly verbose and soulless way. I was hoping for a halfway-decent spelling corrector.
Can anyone help
I do a fair amount of writing and have actually put together several custom GPTs, each with varying degrees of freedom to rewrite the text.
The first one acts strictly as a professional editor—it's allowed to fix spelling errors, grammatical issues, word repetition, etc., but it has to preserve the original writing style.
I do a lot of dictation while I walk my husky, so when I get back home, I can run whisper, convert the audio to text, and throw it at the GPT. It cleans it up, structures it into paragraphs, etc. Between whisper/GPT, it saves me hours of busy work.
The other one is allowed to restructure the text, fix continuity errors, replace words to ensure a more professional tone, and improve the overall flow. This one is more reserved for public communique such as business related emails.
"Bring your own key" has the same amount of syllables as "BYOK"
I'm not a native speaker, and the nice thing about this approach is that I seem to be learning to write better instead of just delegating the task to the machine.
My partner and I were just talking about how useful that would be, especially driving in the car when all of the "we should..." thoughts come out of hiding. Capturing those action items more organically without destroying the flow of the conversation would be heavenly.
> It's designed to be a more customizable and privacy-respecting alternative to Grammarly.
> This extension requires an OpenAI API key to function
I disagree with this description of the service
No, it's not an "Open Source alternative to grammarly", it's an OpenAI wrapper
Meet Harper https://github.com/elijah-potter/harper
Speaking for myself, I clicked on this thinking it might be open source in the sense of something I can run fully locally, like with a small grammar-only model.
Imagine writing a shell script that cuts and converts video by calling ffmpeg, would you say it was “a video converter written in bash”? No, the important part would not be in bash, that’s just the thin wrapper used to call the tool and could be in any language. Meaning it would be useless to anyone who e.g. worked on a constrained system where they are not allowed to install any binaries.
Same thing here. If you only run open-source software for privacy reasons, sending all your program data to some closed server you don’t control doesn’t address your issue. There’s no meaningful difference between making an open-source plugin that calls an OpenAI API and one that calls a Grammarly API.
This repo is _only_ really 7 sentences, like "Please correct spelling mistakes in the following text: " (these https://github.com/zlwaterfield/scramble/blob/2c1d9ebbd6b935...)
Everything else is uncreative, and possibly un-copyrightable, boilerplate to send those sentences to OpenAI.
All of the creative software happens on OpenAI's servers using proprietary code.
If I sold you an electrical generator, but the way it worked was by plugging it in, would you say it's fair to say it's a generator?
You can have a look at the screenshots in the repository or on the store page.
They've been reasonably transparent about how things work, e.g. this blog post from 2018: https://www.grammarly.com/blog/transforming-writing-style-wi...
It worked extremely well, as you say I think by using basic rules engines.
I’ve canceled my subscription recently as found it getting worse, not better, I suspect because they are now applying LLMs.
The suggestions started to make less sense and the problem with LLM suggestions is all your writing takes the tone of the LLM, you loose your personality/style in what you write.
The basic rules approach worked much better for me.
I'm not sure what kind of AI Languagetool uses but it works really well!
https://github.com/languagetool-org/languagetool/blob/master...
Kind of a shame it says it’s specifically for Chrome then. Where’s the love for Firefox?
All of that to say, this is of course a great addition to the ecosystem.
Currently, it's only for Mac, but I'm working on an Electron version too (though it's quite challenging).
Check out https://steerapp.ai/
If this tool really wants to compete with Grammarly.
Also, to provide some feedback, it would be awesome to make it automatically appear on the text areas and highlight errors like Grammarly does, it creates a much better UX.
Obviously not important enough to put in the title, or a submission statement here, though. Curious.
You don't get the AI based extras like paraphrasing, and the other bits listed in as premium only (https://languagetool.org/premium_new), but if you install the n-gram DB for your language (https://languagetool.org/download/ngram-data/) I found it at least as good as, for some examples better than, Grammarly's free offering last time I did a comparison.
Thanks.
You also can self-host and we do that at my workplace, because we deal with sensitive documents.
I looked into the Scramble code[0] and it seems there are few pre-defined prompts(const DEFAULT_PROMPTS).
[0] https://github.com/zlwaterfield/scramble/blob/main/backgroun...
Sorry, but we have a fundamental disagreement on terms here. Sending requests to OpenAI is not 100% local.
The OpenAI API is not free or open source. By your definition, if you used the Grammarly API for this extension it would be a 100% local, open source alternative to Grammarly too.
English can be a very nuanced language - easy to learn, difficult to master. Grammarly helps with that.
The advantage is not spell checking. It is grammar and style improvements. It tells you things like "this language is informal", or "this is a better word for that".
For example, most Slavic languages don't have the same definite/indefinite article system English does, which means that whilst someone could speak and write excellent English, the correct usage of "a" and "the" is a constant conscious struggle, where having a tool to check and correct your working is really useful. In Greek, word order is not so important. And so on.
Spell check usually just doesn't cut it, and when it does (say, in Word), it usually isn't universally available.
Personally, I have long wanted such a system for German, which I am not native in. Lucky for me DeepL launched a similar product with German support.
A recent example for me was that I was universally using "bekommen" as a literal translation of "receive" in all sentences where I needed that word. Through DeepL I learned that the more appropriate word in a bunch of contexts is "erhalten", which is the sort of thing that I would never have got from a spell check.
Grammarly is notably a Ukrainian founded company.
---
That's super useful for people who run say ollama with an nginx reverse proxy in front of it (that adds authentication).
Maybe sending some emails, writing or proofreading some docs -- what you'd do in a business day
Then again, there are people who genuinely believe they could trivially rewrite curl.
https://daniel.haxx.se/blog/2021/05/20/i-could-rewrite-curl/
And some people just try to be polite and it only costs a couple tokens.
I suspect it would be just as easy to write a paper that saying please has absolutely no effect on the output. I feel like gpt4 is/was stochastically better on some days and at some hours than others. That might even be wrong though too. The idea that it is provable that "please" has a positive effect on the output is most likely a ridiculous idea.
> But if I had to get into the underlying structure, I could.
While I don't think they're completely useless (though its close), calling them fantastic replacements feels like an egregious overstatement of their value.
EDIT: Also wanted to note that I think this becomes as much an expectations-setting exercise as it is evaluation on raw programming performance. Some people are incredibly impressed by the ability to assist in building simple web apps, others not so much. Experience will vary across that continuum.
Deep seek lite was essentially useless. Too slow and too low quality edits.
I’ve been programming for about 17 years, so the things I want aider to do are a little more specific than building simple web apps. Larger models are just better at it.
I can run the full deepseek coder model on some cloud and probably get very acceptable results, but then it’s no longer local.
The fact that you can’t is the point of the comment. You could get into the underlying structure of other things, like the C interpreter of a scripting language.
It just so happens to be that a lot of useful stuff is in that box, and LLMs are handy at bringing it out in context. Getting them to “think” is tricky, and it’s best to remember that what you are really doing is trying to get them to talk as if they were thinking.
It sure as heck isn’t programming lol.
Also, it’s useful to keep in mind that “hallucinations “ are not malfunctions. If you were to change parameters to eliminate hallucinations, you would lose the majority of the unusual usefulness of the tool, its ability to synthesise and recombine ideas in statistically plausible (but otherwise random) ways. It’s almost like imagination. People imagine goofy shit all the time too.
At any rate, using agentic scripting you can get it to follow a kind of plan, and it can get pretty close to an actual “train of thought”facsimile for some kinds of tasks.
There are some really solid use cases, actually, but I’d say mostly they aren’t the ones trying to get LLMs to replace higher level tasks. They are actually really good at doing rote menial things. The best LLMs apps are going to be the boring ones.