Still WIP
This commit is contained in:
spushk
parent
2109aaa16d
commit
6d28a2f53a
GPG key ID:
7797B761E105E8AA
1 changed files with 690 additions and 383 deletions
117
README.md
117
README.md
|
|
@ -1,5 +1,28 @@
|
|||
# firaid-moz
|
||||
> I don't know why I named it this way
|
||||
|
||||
> [!NOTE]
|
||||
> This software is provided under terms of BUSL 1.1 license in most permissive mode,
|
||||
> with exceptions for some automation.
|
||||
> Please, consult the [LICENSE](LICENSE.md) if you're about to perform AI-assisted analysis
|
||||
|
||||
---
|
||||
|
||||
**Disclaimer:**
|
||||
|
||||
The software provided as-is, without warranty of any kind, express or implied.
|
||||
This project is an educational proof-of-concept experiment.
|
||||
|
||||
To clarify:
|
||||
|
||||
- I am not responsible for your judgement on content origin (AI/human/assisted)
|
||||
- I am not responsible for your attitude towards AI/human/assisted content
|
||||
- I am not responsible for any consequences of such a judgement
|
||||
|
||||
Classifications performed by this extension are not based on private or social opinions. This extension does not rely on reputations, lists, filters, etc.
|
||||
|
||||
The only method used by this extension to determine origin of content is a bunch of linear algebra in a black-box, aka "ML model", which decisions may or may not be biased - depending on training dataset, which is out of scope of this project.
|
||||
|
||||
---
|
||||
|
||||
A simple local webextension, which can run a bunch of small ML models to infer some properties of DOM content, such as:
|
||||
|
||||
|
|
@ -8,61 +31,93 @@ A simple local webextension, which can run a bunch of small ML models to infer s
|
|||
|
||||
## AI/Human classification
|
||||
|
||||
> **NONE** of ML/AI/whatever models is 100% accurate by the very fundamental design. Any ML model *pretends* to do some job with certain degree of accuracy. So this extension *pretends* to know which content is AI-generated and which is not - with a certain degree of accuracy.
|
||||
> **NONE** of ML/AI/whatever models is 100% accurate by the very fundamental design. Every "AI" simply *pretends* to posess some information. So this extension *pretends* to know which content is AI-generated and which is not - with a certain degree of accuracy.
|
||||
>
|
||||
>If you don't completely trust the "AI slop" - you should not completely trust this extension also!
|
||||
> If you don't trust the "AI slop" - you should not completely trust this extension also!
|
||||
>
|
||||
> False-positives? - Possible
|
||||
> False-positives - Possible
|
||||
>
|
||||
> False-negatives? - Probably
|
||||
> False-negatives - Probably
|
||||
>
|
||||
> This extension - like **every other AI** - is a **tool** which aims to help you and **not** replace your own discretion.
|
||||
> User discretion - Highly advised
|
||||
|
||||
No, I do not hate the AI. It's cool. I hate the *slop* - when the entire "technical guide" article is nothing more then a bunch of AI ejacuation. But how do I know that slop is a slop?
|
||||
I hate the *slop* - when the entire "technical guide" article is nothing more then a bunch of AI ejacuation. But how do I know that slop is a slop?
|
||||
|
||||
Yeah.. I have to read it. And I had _enough_ of reading. So I want to see a warning badge like "its AI" - to help me navigate through text/image content.
|
||||
|
||||
AI/Human classification can be done in four different modes (sorted in order of preference for the potato hardware):
|
||||
AI/Human classification can be done in three modes:
|
||||
|
||||
0. _(potato)_: No classification
|
||||
1. _(normal)_: Lightweight model, which can have "no decision at all" in some cases
|
||||
2. _(high)_: Lightweight model still does its job, but when its decision is weak - heavy model takes the task
|
||||
3. _(ultra)_: Heavy and more accurate model performs always, lightweight model is disabled
|
||||
1. _(lite)_: Lightweight model
|
||||
2. _(mixed)_: Lightweight model runs first, but when its decision is weak - heavier model takes the task. This is usually faster, but requires both lite and heavy models to be loaded thus eating lots of RAM
|
||||
3. _(hard)_: Heavier and more accurate model takes all tasks, lightweight model is disabled
|
||||
|
||||
The _normal_ mode requires ~150Mb of RAM for text classification and ~200Mb of RAM for image classification.
|
||||
The _lite_ mode requires ~80Mb of RAM for text classification
|
||||
|
||||
The _ultra_ mode requires ~700Mb of RAM for text classification and ~500Mb of RAM for image classification.
|
||||
The _mixed_ mode requires ~380Mb of RAM for text classification
|
||||
|
||||
## `Microslop` mode
|
||||
The _hard_ mode requires ~300Mb of RAM for text classification
|
||||
|
||||
> Slop-sloppity slop, micro slop slop slop, microslop microslop microslop
|
||||
## Installation
|
||||
|
||||
That's easy. Turn is on and every time the page contains word "microsoft" (except for the URLs) - this extention will replace it with "microslop"
|
||||
### Building from source
|
||||
|
||||
## `Githief` mode
|
||||
This webextension can be built from source. The following instrumentation is required:
|
||||
|
||||
You know what they do with your "licensed" and "private repository" code.
|
||||
- `cargo` + `cargo-make` + `wasm-pack` (cargo-make will install it automatically)
|
||||
- `pnpm`
|
||||
|
||||
Turn this on - and all "github" (except for urls) shall turn into "githief"
|
||||
To build both firefox and chrome release versions, run the `release` task:
|
||||
|
||||
## `Хуяндекс` mode
|
||||
```bash
|
||||
cargo make -p production release
|
||||
```
|
||||
|
||||
At 2025 the "yandex" corporation was caught in bypassing the data permissions on Android smartphones to collect and possibly provide all kinds of your personal information to the third-parties; to track you; to deanon you; thus violating your consent and privacy.
|
||||
If you do not have `wasm-pack` (cargo) installed, cargo-make will install it for you.
|
||||
|
||||
You can read more about it [here](https://localmess.github.io/)
|
||||
#### Models
|
||||
|
||||
They stopped that practice after disclosure, yes, but they did it nontheless.
|
||||
To only build models, clone the repo, and run `models` task within `production` environment:
|
||||
|
||||
So you can turn on this feature and every occurence of word "яндекс" shall be replaced for "хуяндекс" (russian wordplay, something like "fuck-yandex") and every occurence of word "yandex" (except for urls) shall be replaced with your custom word (I did not figure out proper wordplay for it yet).
|
||||
```bash
|
||||
cargo make -p production models
|
||||
```
|
||||
|
||||
---
|
||||
This will compile all implemented models into WASM packages, which are later used by webextension.
|
||||
|
||||
## A bit of technical details
|
||||
NPM-ready packages for both ai-text models will be located under `<repo>/wasm/pkg` directory
|
||||
|
||||
## Contributing
|
||||
##### Using different models for inference
|
||||
|
||||
Wooow, I am so very glad thay you're reading this section! Thank you in advance for interest to contributing.
|
||||
You can build the [text classification](./wasm/model-ai-text) with any ONNX model from huggingface.
|
||||
|
||||
As you can see - this is not a github, but there is a [mirror](https://github.com/yiffyrusdev/spushks-firaid-moz), which my friend was so kind to provide. You can open issues and make PRs there - we'll merge them, preserving your identity.
|
||||
Following env variables are required:
|
||||
|
||||
Also, if you sumbit any code, it is considered Apache-2.0
|
||||
- `_MK_MODEL_AI_TEXT`: name of the huggingface repository (e.g. "onnx-community/text-ai-detection")
|
||||
- `_MK_MODEL_AI_TEXT_ONNX`: name of the onnx model file under the repository (e.g. "onnx/model.onnx")
|
||||
|
||||
Following env variables are optional:
|
||||
|
||||
- `_MK_MODEL_AI_TEXT_ATTENTION`: set to `yes` or `1` if your model uses attention mask
|
||||
- `_MK_MODEL_AI_TEXT_TYPES`: set to `yes` or `1` if your model uses token types tensor
|
||||
|
||||
Not every model will work, and not every model which works will compute valid results.
|
||||
Major constraints are:
|
||||
|
||||
- FP32 or FP16 weights (quantizations, int8, etc will be available when implemented by burn/burn-onnx)
|
||||
- Sane size, 500Mb is your soft ceiling (browsers would likely refuse to load enormous WASM)
|
||||
- Two label classification, where label `[0]` is Human and `[1]` is AI
|
||||
- Output is `[0, 2]` sized tensor of *logits*
|
||||
|
||||
If you encounter compilation error, and compiler complains about invalid `forward` function arguments,
|
||||
most likely you set the wrong `_MK_MODEL_AI_TEXT_ATTENTION` and `_MK_MODEL_AI_TEXT_TYPES` values.
|
||||
|
||||
#### Extension
|
||||
|
||||
> You must build models before running this
|
||||
|
||||
To only build the extension, clone the repo, build models, then you can run following tasks:
|
||||
|
||||
- `extension`: build and zip firefox and chrome extensions
|
||||
- `extension_moz`: only build firefox extension
|
||||
- `extension_chro`: only build chrome extension
|
||||
- `dev`: run wxt dev server (chrome variant)
|
||||
|
|
|
|||
Loading…
Reference in a new issue