Discoveries

I needed access to Nvidia GPUS to play around with Vision Embedding (specifically ColPali).

The document i wanted to index had few thousand pages and was not possible to do via my MAC.

Was looking around and found this platform called Prime Intellect. I was able to rent a GPU ($.37 /hr for 4090) and setup Jupyter Notebook in roughly 3 mins

Found this link where @stirckvl has compiled how llms is used in real world. contains links to various case studies

Find the link here

Interactive image editing system that lets you edit images with three smart brushes - add new elements, remove objects, or change colors. Just draw strokes and the AI understands your intent without typing prompts.

Look at example below

And you can try it here with your own images

Super fast transcription tool optimized for Apple Silicon Macs using MLX framework. Install dependencies first, then you can convert any media file and transcribe it.

# Install dependencies
brew install ffmpeg
git clone https://github.com/JosefAlbers/whisper-turbo-mlx.git
cd whisper-turbo-mlx
pip install -e .

# Convert media to wav
ffmpeg -i input.mp4 output.wav
# Transcribe
wtm output.wav --quick=True

Source: JosefAlbers/whisper-turbo-mlx