OpenAI – the company behind popular AI language model GPT-3 and ChatGPT has finally announced the newest version of its AI model – GPT-4.
GPT-4 was announced in a live event.
Key Highlights of GPT-4
- GPT-4 is much more capable compared to GPT-3.5/ChatGPT.
- GPT-4 is multimodal. That is, it supports images as input and gives appropriate output.
- Supports 25,000 words context
- Being tried and tested by many partners: Duolingo, Be My Eyes, Stripe, Morgan Stanley, Khan Academy and even Government of Iceland!
- Available to try out for ChatGPT Plus users and there is waitlist for API access – Join API waitlist
According to OpenAI, GPT-4 is their most advanced AI system in producing safer and useful responses. In short, GPT-4 is more creative, able to handle complex context is more accurate way.
In addition to being more capable, GPT-4 supports images as input and give relevant output (Image feature is still in preview and not available publicly). It also now has the capability of 25,000 words context, this is again a massive improvement over GPT-3.5
The company says GPT-4 shows great improvements in practical and AI benchmarks. Like the Uniform Bar Exam, LSAT, SAT Math, and SAT Evidence-Based Reading & Writing exams.
You can check in more detail about all the exams and benchmark performance of GPT-4 in their detailed research article here.
The company said GPT-4 was already ready months back, but they delayed the launch, focusing more on safety. OpenAI stressed that the system had gone through six months of safety training internally.
Availability of GPT-4
The GPT-4 is generally available for use for ChatGPT Plus. Keep in mind, the image feature is still in preview and currently not public.
GPT-4 is not available for general ChatGPT users. OpenAI said the GPT-4 will be available for everyone after some time. So, free users may have to wait some more time to use GPT-4.
To get API access of GPT-4, you will have to join a waitlist, as the API for GPT-4 is currently not available. You can join waitlist here. Recently, OpenAI announced the general API availability for GPT 3.5 along with Whisper speech-to-text technology.
However, the system is still not fully mistake-free. For start, GPT-4 does not have any awareness and knowledge of events that have occurred after (September 2021).
Additionally, Microsoft which revealed that its Bing chatbot which had ChatGPT integration has been powered by GPT-4 all along.
Many, and many users were able to jailbreak Bing’s Chatbot safeguards in multiple creative ways, getting the bot to offer dangerous advice, threaten users, and make up information.
GPT-4 seems quite an update from ChatGPT. We will be seeing a lot more updates coming from GPT-4 and OpenAI.
You must log in to post a comment.