What’s the news: OpenAI makes the news once again, as it releases its latest generation of AI language model ‘GPT-4’ for ChatGPT Plus users. According to the company’s post, “GPT-4 is more reliable, creative, and able to handle much more nuanced instructions than GPT-3.5.”
What can GPT-4 do?
GPT-4 was trained on Microsoft Azure AI supercomputers. As per the blog, it passed a simulated bar exam with a score around the top 10 percent of test takers, whereas GPT-3.5 scored around the bottom 10 percent of test takers.
Response to visual inputs: GPT-4 can accept a prompt of images, not just text. Imagine you set-up your laptop for a presentation and your device won’t connect to the projector for some unknown reason. Rather than trying to narrate the entire problem, you can send your query along with a photo of the devices and ask GPT-4 for a solution. It will then respond in text.
“Image inputs are still a research preview and not publicly available,” said the post.
Personalise GPT-4’s personality: Users can also shape the tone, style, speech, and tasks of the AI by describing those directions in the “system” message. For example, you can instruct the AI to always encourage a student to find the correct answer to a problem but never give the answer to the student. OpenAI mentioned that this feature requires improvements, particularly because “system messages are the easiest way to “jailbreak” the current model, i.e., the adherence to the bounds is not perfect.”
Other than this, developers say that GPT-4 has shown better ‘impactful’ performance for functions like support, sales, content moderation, and programming.
STAY ON TOP OF TECH POLICY: Our daily newsletter with top stories from MediaNama and around the world, delivered to your inbox before 9 AM. Click here to sign up today!
What are GPT-4’s limitations?
Like previous models, GPT-4 “hallucinates” facts and makes reasoning errors, as per OpenAI’s post. It also lacks knowledge of events after September 2021 and does not learn from its experience. Further, it can sometimes make simple reasoning errors or be extremely gullible in accepting obvious false statements from a user. It can also fail at hard problems such as introducing security vulnerabilities into the code it produces. Another issue is that GPT-4 can be confidently wrong in its predictions as it does not double-check its work.
Who can use GPT-4? The AI is available on ChatGPT Plus, a premium use service, and as an API (Application Programming Interface) for developers to build applications and services.
“We are releasing GPT-4’s text input capability via ChatGPT and the API (with a waitlist). To prepare the image input capability for wider availability, we’re collaborating closely with a single partner to start,” said OpenAI.
Why it matters: In just a few months, ChatGPT and its many versions have been noticed by big companies like Microsoft, while companies like Google are working to create rival AI models. The popularity is such that even the Indian government is working on a ChatGPT-powered WhatsApp chatbot to answer questions. Considering the ways this technology is making headway and the speed at which it is being picked up by corporations, it is important to understand the AI’s various versions, capabilities, and limitations.
This post is released under a CC-BY-SA 4.0 license. Please feel free to republish on your site, with attribution and a link. Adaptation and rewriting, though allowed, should be true to the original.
Also Read:
- After Artists And Coders, News Outlets Up Against OpenAI For Using Their Articles To Train ChatGPT
- Explainer: What Does It Mean To “Jailbreak” ChatGPT
- Explainer: Meet The New ChatGPT-Powered Microsoft Bing Search Engine And Edge Browser
- Beware, Gaming Journalists; ChatGPT Has Quite A Lot To Say About Online Gaming In India
