When 2M token context? GPT-4.5

@GaryMarcus @MMikeMMa @somewheresy @nvidiaomniverse @mishadavinci @OpenAI

9/ If you found this thread informative, please consider retweeting it to spread the word and learn more about the rise of transformers in AI!

8/ That being said, the continued development and advancement of tranformers is certainly worth keeping an eye on as we continue our journey towards general AI.

7/ In order to reach true general AI, we need systems that can perform a wide range of tasks at a human- or even superhuman-level. Tranformers have certainly made progress in this direction, but there is still much work to be done.

Perspectives on General AI: 6/ But while tranformers are certainly making strides in advancing AI, we are still a long way from achieving the ultimate goal of general artificial intelligence.

5/ It is likely that future models will incorporate the best aspects of both transformers and traditional algorithms such as convolutional neural networks.

4/ While transformers have the potential to greatly improve AI tools, their high computational power requirements in the training phase may be a barrier to widespread adoption.

3/ Researchers are also exploring the potential of transformers for multimodal processing tasks, in which multiple types of data are combined and analyzed, and for use in various other AI domains including robotics, speech analysis, and protein structure prediction.

2/ One of the key features of transformers is "self-attention," which allows them to connect every element in the input data to every other element, enabling them to quickly process large data sets and achieve high accuracy in tasks such as image classification.

1/ Transformers are a type of artificial neural network that have revolutionized language processing tasks in AI, and are now being explored for use in other areas such as computer vision and image generation.

The Rise of Transformative AI: Tranformers are pushing towards General AI (🧵🧵)

I just subscribed to The Road to AI We Can Trust

It made me think it will certainly be a great tool -it codes in Python! and it works!- but has to be used with care (you cant just blindly trust what it "tells" you)

Weak: has no grips on reality. Talks on "made up" conclusions, as if they were scientific true. There, I had some concerns...

What I felt wile using/testing GPT, is that it is not knowledge based (but it is so nice!) It seems it is made to replicate human like speach. Strong point is that, it seems it can be very imaginative -creatting(?) inventing stories-

3/ @GaryMarcus just published a paper: "The Next Decade in AI: Four Steps Towards Robust Artificial Intelligence" (the arXiv link if you got curious:

2/ @GaryMarcus raised interesting points. I keep thinking about them. I have tested chatGPT in my science field. Good tool, but need to use with great care (lots of hallucination : )

1/ Great discussion about current stage of AI. (🧵🧵) It is provocative in some sense (gets you thinking about cognition and about how current DL approaches might be missing something --> Therefore, could be improved) https://t.co/9IhYYBXAb3

#ICP The Internet Computer