Recently, GPT-4, the latest iteration of OpenAI’s model language, was released.
The GPT-3.5 language model served as the foundation for ChatGPT, the chatbot that saw an unparalleled surge of popularity around the world. A new upgrade from OpenAI today allows for the addition of some very spectacular new features while also improving the machine’s existing capabilities.
The new restriction is 25,000 words.
It was inevitable that this announcement would be heard around the world. The creator of ChatGPT, OpenAI, has lately made available the GPT-4 version of their brand-new template language. A brand-new item that is now only accessible to a certain segment of the public due to the requirement that you be a member of the ChatGPT Plus program.
Yet, a whole new universe becomes accessible to individuals on the VIP list. The AI’s capabilities have significantly expanded, starting with the prompt level (the text entered as a request within the chatbot). GPT-4 can now evaluate user-proposed texts with up to 25,000 words in length!
The picture follows the writing!
Nonetheless, GPT-4 image recognition will undoubtedly be the function that receives the most attention. GPT-4 will be able to reply to prompts that involve words and graphics and respond after assessing them thanks to the partnership with a start-up called Be My Eyes. Greg Brockman, a co-founder of OpenAI, summarizes, “It can flexibly take data that randomly intersperses pictures and text, much like a manuscript.
In order to show the growth of the model language, which has become “multimodal”, the latter inserted into the machine a drawing defining a website project. In response, GPT-4 generated a thorough answer outlining the actions to take in order to reach this site. This feature, unlike the rest of the new version, is not yet accessible and is still being evaluated within Be My Eyes.
Many partners are already employed.
Yet, agreements with OpenAI have grown in order to employ GPT-4 even without having access to this technology, which at first look has everyone’s attention. Hence, in order to satisfy the support requirements of its business clients, online payment expert Stripe has included the language model into its services. AI will be used to identify possible fraudsters as well.
The well-known language-learning program Duolingo will soon launch a new subscription called Duolingo Max that will include GPT-4. Although the app has always employed AI, GPT-4 should enable it to advance significantly. Conversational practice and contextual feedback on errors were holes in the learner experience that needed to be filled, according to product manager Edwin Bodge.
Enough to add a bit more cash to the business’s coffers. So even though it receives assistance from the company Bill Gates created, the technology it is creating has unusually high operational expenses.
Safer and factually correct
Lastly, GPT-4 has been made more “secure,” which was undoubtedly a focus at OpenAI in light of the numerous issues surrounding chatbots’ propensity for spreading false information. A task that, according to OpenAI, would have taken six months to complete and that enables Microsoft’s partner firm to provide findings of higher quality.
In order to react to requests for illicit content, such as coding malware, GPT-4 would be 82% less probable than GPT-3.5. The machine’s accuracy has also been improved, as it is now 40% more likely than in the past to provide a factual response to a query.
Unfortunately, the AI’s propensity to “hallucinate” when faced with a challenge has not yet been totally overcome. In support of the last cautionary statement from OpenAI, “Extreme care should be exercised when utilizing the outputs of a language model, especially in high-stakes scenarios,”
