I was at an event with 4 speakers, where they talked about AI, mostly in computer vision. Here is what stood out as relevant to me.
- Hugging face, described by the speaker described as “the github of AI”, holding many models. (Apache2 license means free to use).
Computer vision stuff:
- They talked mostly of how AI can be used to navigate GUI and analyze pictures. Qwen-2.5 VL is the SOTA for vision language models. Bagel looks awesome.
- With 5 images of ourself looking straight at camera and 20 min of training you can make yourself be anywhere! He mentioned inference providers that allow using multiple models (like we do).
Some links he gave to learn more:
hf.co/learn
https://github.com/merveenoyan/smol-vision
I also met with a friend who is in AI, he just did a logging feature for an app with 2 million ppl, and is also using cursor - he’s been freelancing for years. So I asked him for tips.
He said that it’s a good idea to write the tests manually and start from that so that AI can have free reign without breaking things. He takes a lot of time working on the architecture.
Finally he shared this youtube channel to stay updated on AI: https://m.youtube.com/@theAIsearch