There is news of ChatGPT4 and rumors have it that the new ChatGPT shall bring about significant improvements to OpenAI’sChatGPT. GPT stands for Generative Pre-trained Transformer (GPT), it is a text generation deep learning model that is trained on the internet information. GPT is used for questions and answers, code generation, text summarization, machine translation, and classification.
Although GPT-4 has not yet confirmed any details about GPT 4, it has been confirmed that it is in the pipeline.
What’s GPT-4?
OpenAI is creating a new language model that will generate text similar to human speech. It will feature more sophisticated and advanced technology than the previous ChatGPT model. Sam Altman, OpenAI’s CEO, recently confirmed the rumors regarding the launch of the GPT-4 model.
We shall discuss the given information and merge it with the current trends to give an estimate of the model size and optimal parameters.
Model Size
Altman stated that GPT-4 is not going to be much bigger than GPT-3. So, it is predicted that it is going to have about 175B-280B parameters, like Gopher, by Deepmind.
Text-only
The new GPT-4 is going to be a text-only model. Altman said that it will not have multimodal capabilities like DALL-E. The reason is that multimodal is more complex to build than vision/language only. Merging visual and textual information is complicated as well as challenging. Also, they should have improved performance than DALL-E 2 and GPT-3. So, GPT-4 is not going to feature anything fancy.
Better Code Generation
Rumors suggest that GPT-4 is going to have better computer code generation and can generate both images and text.
AI Alignment
Compared to GPT-3, we can expect GPT-4 to be more aligned. At the moment, OpenAI is having a hard time with AI alignment. The company aims for language models that follow user intent and stick to values. The first step has been taken as InstructGPT is trained. It happens to be a GPT-3 model instructed on human feedback to follow directions.
GPT-4 Expected Release Date
Nothing can be said with surety regarding the release date of GPT-4. However, as per New York Times, we can expect it to arrive in the first quarter of this year. Since it’s March, the launch can be expected in a few weeks. On the other hand, Microsoft has confirmed that GPT-4 would arrive on the week of March 13 and would be a multimodal model capable of generating text, images, and video too.
Wrapping Up
GPT-4 is going to offer better performance on a similar size as GPT-3. Moreover, it shall be more aligned with human values and commands. You might hear conflicting news regarding GPT-4 as well suggesting that it’s going to be a text-only model with 100 trillion parameters that will emphasize code generation only. There are many facts we are unaware of yet, as plenty of rumors are circulating. So, everything is speculation at this point.
However, GPT-4 will be less biased, more secure, and more accurate, not to mention cost-efficient.
