Thing 1 - This AI can read your mind (Yes, for real)
No, you can't download and use it. Yet.
It requires scanning you in the fMRI (functional Magnetic Resonance Imaging) machine and training the system on your brain. Last I check those ain't cheap.
But Kernel is a thing... 🤔
They gathered brain scan data from four people who looked at 10,000 different pictures (that's like 10M words if you know what I mean).
They created their LDM (Latent Diffusion Model), which is like a smart algorithm that can generate images from just a text description. They also did some decoding and encoding analyses, to translate brain signals into something the LDM can understand and predict.
Here's the exciting part
They found that their method can make super high-resolution images from fMRI signals.
They also took a deep dive into the LDM's insides, linking them to brain regions and showing how the magic of image-to-image transformation happens.
This is bound to improve so you might as well start thinking about cool ways you'd use this new interface for.
Method potential downsides (optional)
- relies on a pre-trained latent diffusion model that may not generalize well to other image domains or text descriptions.
- assumes a linear mapping from fMRI signals to latent representations, which may not capture the nonlinear and complex relationships between brain activity and visual stimuli.
- does not account for individual differences in brain anatomy and function, which may affect the quality and accuracy of image reconstruction.
- does not evaluate the robustness and reliability of image reconstruction across different sessions, stimuli, and noise levels.
Thing 2 - AI updates of the week
You seemed to have liked it last week. But also, I can't look away 👀
- Box announced Box AI
- Mojo🔥 - a new programming language for ML that is high-performance and backward-compatible with Python
- Harvard webinar - How to use AI tools to make teaching easier and more effective
- StarCoder, a free alternative to code-generating AI systems (competition to Github's Copilot)
- Geoffrey Hinton (Godfather of AI) - quits Google to warn about the dangers of AI
It's as if you had 10,000 people and whenever one person learned something, everybody automatically knew it. And that's how these chatbots can know so much more than any one person.
- Geoffrey Hinton
Thing 3 - You're Your Own Best Friend – Make the most of it
Just as one person delights in improving his farm, and another his horse, so I delight in attending to my own improvement day by day.
- Epictetus (quoting Socrates)
You are keeping yourself company 24/7.
You are the person you spend the most time with.
Why wouldn't want you to make that person better?