Emergent abilities in LLMs like ChatGPT
But this just caught my fancy and as I was reading this and it just was so amazing to read about emergent properties of large language models and AI and compared and think about emergence in other parts and other scientific systems as well. And it just makes for a really interesting kind of comparison discussion. So if you have any thoughts on this, if this is the kind of stuff that you enjoy reading or talking about, feel free to hit reply and let me know. Bye
Shammi Mohamed
@shammi · 4:49
But what is fascinating over here is that GPT works under the skin by just very one simple fundamental building block, which is really a prediction of the next line of text. And it is fascinating that something as simple as the autocomplete for your phone text, it's the same thing that is scaled up in terms of the number of parameters that is fed into it, or rather the scale of the neural network that is used to deduce the next prediction is exhibiting these emergent properties
Shammi Mohamed
@shammi · 4:50
That is the only goal of this system. You will start to see more emergent properties, emergent goals that will satisfy this core goal of being able to predict things better with any person you speak. And all of a sudden right now you have a system that is growing like to be more self aware and more it has more goals if we allow it to be free in our society
But my point is, even if you're able to replicate the human way of thinking and building a neural network which is much more aligned to how we learn something, right? You would still not get consciousness from that. You will still not get sentience from that. Because that is emerging from a whole different part of what it means to be human, what it means to be alive. And that's a biological part of it
Shammi Mohamed
@shammi · 4:48
And perhaps that's why we say warm blooded animals versus cold blooded animals metaphorically. That is the experiences that make us feel one way are not the same for other animals, especially reptile reptiles, right, and other animals who don't care for their young quite the way mammals do. So likewise I feel like in this case it's a little different, right?
Swell Team
@Swell · 0:15
Shammi Mohamed
@shammi · 4:52
It's going to create these centers of processing because of emergence, things that handle vision well, things that handle certain things well, and things that can combine and get a creative response out of it. But eventually it will take all of that and spit out its output to us. And we will and our responses will give it a high, or rather a prediction score high, and it will keep going towards that continuing in another as well
Shammi Mohamed
@shammi · 3:44
And they do not have consciousness of the same order of what we do because what eventually pleases us is not what pleases them in many cases. Likewise, if you want to look at this as another entity that kind of thinks for itself, and the reason it thinks for itself will be given the capability to what do you call improve itself, basically, you give it an entire substrate and allow it to program itself, which we are going to do it because that's the way, like you want
Shammi Mohamed
@shammi · 1:03
And number two, I in my opinion, or rather, I think you can look at it in another way, DNA is probably the best form of software compression there is. It compresses the code to create an entire organism as the human being in 23 chromosomes. And that is saying something. But at the end of the day, it is all software. But cheers. This was awesome. Thanks for allowing me to participate in this. And yeah, we should continue exploring this more and more