Elon Musk says he will launch rival to Microsoft-backed ChatGPT...
(www.reuters.com)
You're viewing a single comment thread. View all comments, or full comment thread.
Comments (18)
sorted by:
Because Elon has access TB's of data in the form of tweets from a billion people. That is a significant advantage that open source projects don't have.
Remember when Q told us in post 605 about an Internet Bill of Rights? You think that was just so we could post drivel on Fakebook? AI's are a powerful new tool...possibly the most powerful ever made by man. And the algorithm is easy. Getting access to the training data though? That's hard.
My guess? This and the RESTRICT act are the catalysts for it. Watch for some controversy over this announcement by Elon.
I know most people find this very unbelievable, but AI such as ChatGPT are language models, and over training them with, say, billions of tweets of people is actually not effective. It saturates the training and the more you overtrain, the more it starts "hallucinating".
I have tried models that are fraction of the size of ChatGPT that are >90% as powerful as ChatGPT, and can be trained for domain specific data.
ChatGPT was supposed to be this huge leap ahead of the rest that we were never supposed to be able to catch up. Instead, it just made it easy for all of us to reach the same level on our own.
I Elon is focusing on making new models, he is late to the game. He is far better off focusing and research for speeding up the tensor math, for example, or faster algorithms for training/evaluating these models.
Of course, that is only true if you fix the size neural network. If you find your network is overfitting, you can always increase the hidden layers making, in theory at least, a more "intelligent" network. It then becomes a question of whether or not that makes sense for the application at hand. You can never really have too much data. But your point is understood. There has to be a valid reason to build a bigger model.
Sure, bigger models have a reason to be built, but not for general language models like ChatGPT. Sure, if you are gonna build a subject expert or specialized AI doing things we cannot imagine today, you might have to build bigger models.
Billions of twitter conversations, on the other hand, will only yield a generalized LLMs and an overkill at that.
Putting Elon aside, I think some of the emerging tech will be in reducing the current models significantly and making them run on consumer mobile devices. That would be fun.
Think about driving your girl on a road trip and she go’s - well you know. All of that is photographed- and or recorded and sent back to Tesla. Then pervey IT employees post the best (well you know) on the bulletin in the break room and rate them! I’ll still with my 09’ vehicle.