AYDIN KULAN

I find it more appropriate to research how to run a large model on small hardware, rather than how to run a small model on small hardware.

Please feel free to take a look. https://github.com/tarikkaya/aix

I’d like to mention one more thing: when I envision an AGI model, I’m imagining a design that merges these two images. I think a flat space would require a significant amount of computation to capture the necessary contexts. I also believe that a spherical model would facilitate easier access to different perspectives.

So, how do you plan to use an entity’s motor structures without breaking them down? Any model you design without leveraging these motor structures will be very weak. My suggestion is to use motor structures and a scale when necessary, applying different tones for different perspectives

It seems I’m not the only one who thinks the “What if…” series is a blueprint for AGI. Alternative realities make it possible to access different perspectives, and I’m even sure that combining them in this field will succeed. But the real issue is the feeding loop—i.e. the training loop. How do you make it both easy to train and ensure the neural connections stay solid? You can’t have both at once.

Did you build this model inside a sphere?

We need to address the problems of existing models seeing the big picture; moreover, this issue concerns not only VLMs but also LLMs. Imagine a map. If you look at it very closely, you can’t see the big picture. If you view it from a distance, you can’t see nearby objects.

I definitely think that AI needs to effectively use novels, film scripts, and theater scripts to understand human emotions and thought structures. Not only to define emotions, but also because, in my opinion, we need these kinds of sources within thought frameworks inspired by human imagination and creativity. When we consider the datasets of today’s models, I actually think we’re providing very limited information. Remember that without mistakes there is no truth. This should serve as a fundamental basis for defining many expressions across many workflows.