OpenAI, the makers behind the enormously popular artificial intelligence (AI) powered chatbot ChatGPT, launched its algorithm’s newest version, GPT-4, on March 14th.

The latest version is packed with improvements and new features that promise “safer and more useful responses.” There is, however, a catch – you’ll need to upgrade your free account, and once you do, a usage cap restricts you from fully taking advantage of its capabilities.

GPT-4 is the highly anticipated successor of GPT-3.5. GPT-4 uses the same fundamental architecture as its predecessor but with more training parameters and significantly improved features.

Some updates include safer responses, the capability to accept images as input, extended text interpretation and outputs, factual correctness, and more advanced reasoning capabilities.

GPT-4 shines when you look at its performance on standardized tests. When using its complete feature set, including its ability to interpret images, GPT-4 was able to score in the 90th percentile on the bar exam, an exam for lawyer applicants, compared to only the 10th percentile with GPT-3.5.

Unfortunately, as of this writing, those on a free plan won’t be able to enjoy GPT-4 and its capabilities. You’ll need to upgrade to ChatGPT Plus and pay $20 a month. However, even ChatGPT Plus users have a usage cap for GPT-4, only able to generate 100 responses every four hours.

OpenAI will also be making the GPT-4 API accessible to developers via a waitlist.

What Features to Expect with GPT-4?

As mentioned earlier, significant improvements make GPT-4 significantly better than its predecessor. Here is a closer look at these updates.

Safer and More Aligned

ChatGPT users often complain about harmful and biased generated responses and inaccurate, fictitious responses with the GPT-3.5. But with GPT-4, OpenAI, the makers of ChatGPT, promises that the chatbot’s answers are now “safer and more aligned.”

To be more precise, the latest model’s tendency to generate disallowed content is down to 82%, while factual responses are up 40% more. This means the new model is safer for users overall.

These safety and alignment improvements result from human feedback from experts and ChatGPT users, GPT-4’s capabilities, and lessons learned from previous models.

See and Interpret Images

With GPT-3, you can only input text. In GPT-4, the chatbot is now multimodal, which means you can ask the chatbot to perform text or visual tasks.

You can upload photos (graphs, charts, screenshots, memes, and other images) and ask the chatbot to interpret the image based on what you ask the AI. Amazingly, it can generate a functional HTML page based on a website’s uploaded sketch.

However, this new feature is still unavailable to the public and is currently in research preview.

Improved “Steerability”

One of the most welcome updates with the GPT-4 is its improved “steerability.” This update allows ChatGPT users to change or “steer” the behavior of the chatbot based on the directed prompt, like tone or style, to suit their needs better.

Interpret and Output Longer Text

GPT-3 can only handle around 1,500 words at a time, so it falls short on interpreting and generating longer texts. In GPT-4, you won’t run into this problem because it can now generate over 25,000 words of text – that’s 8x more than its predecessor.

For instance, you can input a Wikipedia URL, and the chatbot will analyze the text, and depending on the request, it can provide you with a summary. This improvement benefits users analyzing data on larger documents or creating long-form content.

Smarter With More Advanced Reasoning Capabilities

GPT-4 also excels in reasoning capabilities by making the responses easy to follow, simple, and short. It demonstrates this by outperforming its predecessor in simulated exams. OpenAI GPT-4 across multiple Advanced Placement (AP) tests and common standardized exams, including the LSAT, SAT, and others. GPT-4 performed significantly better than GPT-3.5 on these tests.

About ChatGPT

San-Francisco-based company OpenAI’s ChatGPT made headlines late last year for generating conversational, human-like responses. Unlike traditional search engines like Google and Bing, ChatGPT users can get an answer in a blink of an eye without ever having to sift through millions of search results.

This convenience and its capability to explain simple to complex topics under the sun without sounding like a “robot” resulted in one billion website visits. The alarming growth of ChatGPT has prompted search engine companies like Bing and Google to introduce their own AI chatbots.

Fact-Checking More Necessary Than Ever

AI tools for writing content have been around for years but are still far from perfect.

Take, for instance, ChatGPT. It is only trained until September 2021, which means it is not up to date with recent events. Even the chatbot itself is fully aware of its limitation and will even recommend referring to news sources for the latest information.

ChatGPT lacks proper citations as well, with some users reporting inaccuracies or hilarious responses.

Even with the launch of GPT-4, OpenAI cites limitations such as “social biases, hallucinations, and adversarial prompts” that still need to be addressed.

There are also issues with inputting sensitive information into the chatbot, with companies like JPMorgan banning their staff from using the AI.

However, that is not to say that AI tools are not useful.

Professionals and students can use them to structure their essays, reports, and more. And with Grammarly and other online grammar checkers, they can spend less time worrying about spelling and grammar mistakes and focus on writing pieces that are accurate and factually correct.

This article was produced by Capitalize My Title and syndicated by Wealth of Geeks.

Thank you for reading! Have some feedback for us?
Contact the 24/7 Wall St. editorial team.