# Fine Tuning AI

{% hint style="warning" %}
**Importance of Tuning:** Fine tuning is a process that helps produce the data you're looking for receiving. Tuning can be trial and error along with your in-game settings.

***Join Our discord:***[ ](https://discord.gg/dFg8smncrR)[***https://discord.gg/dFg8smncrR***](https://discord.gg/dFg8smncrR)
{% endhint %}

<figure><img src="https://753012564-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fgg1E2stc3vw49ViGbucH%2Fuploads%2FFVfJMOyf3CzgY6qYMkrI%2Fsettings.png?alt=media&#x26;token=87e3f327-5a5e-41a3-a0c8-b2df964fe0c6" alt=""><figcaption></figcaption></figure>

## Sens Tuning

<figure><img src="https://753012564-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fgg1E2stc3vw49ViGbucH%2Fuploads%2FdxUlCQ3di9WqDnHjSijf%2Fsens%20.png?alt=media&#x26;token=21f03897-2a26-4dfb-8c7a-b2f8fc812546" alt=""><figcaption></figcaption></figure>

Tuning sensitivity can allow the AI to have a smoother detection if there is a lot of chaos in the scene in the case of your sensitivity being a factor, so Vision Bot provides a tuning option because each user can have a different experience until we have enough (voluntary) users who wouldn't mind submitting detection data for training. Until then, sensitivity. tuning is the alternative to adjusting the speed in which the  AI increments the axis.

## Confidence Tuning

<figure><img src="https://753012564-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fgg1E2stc3vw49ViGbucH%2Fuploads%2FEnlsmPd3h6xnIsypgE0m%2Ftuner.png?alt=media&#x26;token=8417e6f1-e089-413e-acbe-b74673fa8d11" alt=""><figcaption></figcaption></figure>

Tuning confidence is key, this can provide you with more or less data depending on your needs.\
The confidence is self explanatory in the terms that:

* 0 means: apply 0 confidence to detections, so whatever the AI thinks it detects, with no constraints.
* &#x20;means: apply 1.0 confidence to the detections, so do not return any detections unless your 100% confident (which can lead to no data being detected if the current dataset is too small. (which it is))
* 0.01-0.99 means: Apply this confidence and return any detections...

Recommended : .35 (35, In our tests, lets the AI breath but also allows strictness, but will change with frequent updates to the dataset.)\
\
The sweetest spot for the dataset size currently being utilized is no less than .28 (28) or things can get pretty weird if Vision Bot is adjusting your axis and may lead to false positives. So test this feature and find what works best with your PC.

## Dead zones (Coming Soon)

Currently in tests we've conducted, dead-zones aren't such a big issue unless you have it set +10 and you take no small movements, in that case don't worry, we're working on implementing accurate dead zone and anti-dead zone algorithms to prevent dead zones from interfering with The Vision Assist Bot.