OpenAI has finally taken the curtains off of the GPT-4. It’s the company’s latest AI language model that will power the ever-popular ChatGPT, Microsoft Bing chatbot, and other several platforms. The company claims the model is “more creative and collaborative than ever before” and “can solve difficult problems with greater accuracy.” More about OpenAI’s GPT-4 AI language model is in this article.
OpenAI has undoubtedly made a name for itself with the phenomenal success of GPT 3.5-powered ChatGPT. The platform went viral almost instantly after its launch in November 2022; and in just five days it was able to reach one million users, the fastest for any platform to date.
Meanwhile, OpenAI was also readying the next iteration of the GPT language model, the GPT-4. But before the official announcement from the company itself, several hoaxes began to circulate in the market.
One of the most prominent rumors was that the GPT-4 will be 500 times more powerful than its predecessor. And will have the ability to write a 200-page book from scratch in less than a day. Well, OpenAI has launched the GPT-4, keeping those exaggerated rumors at bay.
Still, there are quite a few interesting upgrades, with one of the most significant being the addition of multimodal AI language. That is, it can now accept both text and image inputs. However, the responses to queries will still be in text format.
As seen in the above example, the GPT-4 is able to interpret a meme with proper explanation. It generates text outputs for interspersed text and image inputs. However, the image input is in the preview stage and is currently not available to the public.
In casual conversations though, the GPT-4 offers subtle differences from its predecessor. OpenAI does mention that the new model can handle comparatively complex thresholds with better outputs. But it is not as significant as you might expect it to be. Also, it’s now capable of handling 25,000 words of text, from the previous limit of 3000 words.
Scores above 80 Percentile
Moving on, OpenAI is also quite confident about the AI language model’s intelligence. The GPT-4 manages to score 88 percentiles and above on Uniform Bar Exam, LSAT, SAT Math, and SAT Evidence-Based Reading & Writing exams. The AI language is said to have gone through six months of safety training to achieve this output.
- You might also want to read:
Still not flaw-free
GPT-4 is still not flawless. The company has admitted several known limitations such as social biases, hallucinations, and adversarial prompts. Also, GPT-4 still lacks knowledge about events that happened before September 2021.
The AI language model is available to try from ChatGPT Plus with a paid subscription of USD 20 per month. The company has announced a partnership with Duolingo, Stripe, and Khan Academy for integrating GPT-4 into their products. Besides that Microsoft has confirmed that the new Bing is running on GPT-4 language model. While developers can also join the waitlist for using the API.
- Meanwhile, you can watch our Samsung Galaxy S23 / S23+ review.