GPT-4 release.

Comparing to other flavors available

As predicted earlier, OpenAI released GPT-4 a few minutes ago. GPT-4 is the most advanced model, available to Plus subscribers at this time. According to OpenAI, GPT-4 excels at tasks that require advanced reasoning, complex instruction understanding, and more creativity.

I have provided the differences below between the three types of offerings provided by OpenAI. ChatGPT Plus costs $20 for signing up, and unlike the free version, you can be assured of always having connectivity to the system. The GPT-4 release would be a bonus if you are a Plus subscriber.

Capabilities:

• This version is more creative and collaborative than ever before. It can generate, edit, and iterate along with users on creative and technical writing tasks, such as composing songs, writing screenplays, or learning user’s writing style. This particular aspect of “learning user’s style” is going to be an amazing addition in my opinion.

• GPT-4 can accept images as inputs and generate captions, classifications, and analyses.

• GPT-4 can handle over 25K words of text that will allow long-form content creation, extended conversations and document search analysis. These would be very powerful additions for enterprises. Currently the showcase of this implementation includes Stripe and MorganStanely using their internal documents used to train the system and have the chatbot created for internal usage. Now you can source the content from articles like Wikipedia. Following is an example of providing the link from The Verge article about SVB. I asked the system to summarize the article with bullet pointed highlights. Following is what I got.

  

  In the OpenAI site, it was showcased how GPT-4 performed compared to ChatGPT. GPT-4 is in the 90th percentile among bar test and 99th in Biology olympiad.

  

  What about Safety?

  Per OpenAI, human feedback submitted by ChatGPT users has been incorporated and worked with 50 experts for early feedback on AI safety and security. Since this is a RLHF (Reinforcement Learning with Human Feedback), the continuous updates are expected.

It is very exciting to see the advancements in the AI field at this point of time. As always, companies like OpenAI hopefully will hear the community and keep these systems safe and sound for general use.