OpenAi’s GPT-4: New features and unsolved challenges | Generative ai Course Google Online | Generative ai Tools List | Microsoft Generative ai Tools Github | Turtles AI

OpenAi’s GPT-4: New features and unsolved challenges
Isabella V31 May 2024

 

 


OPENAI LAUNCHES GPT-4 : SIGNIFICANT IMPROVEMENTS BUT LINGERING CHALLENGES

 

In 2023, OpenAi launched " GPT-4," the new version of its AI model, available for developers and companies to create custom applications.

GPT-4 is integrated into the premium version of ChatGPT, accessible via monthly subscription, and promises more accurate responses through pre-trained models.

Despite the improvements, GPT-4 retains some of its predecessor’s shortcomings, such as " hallucinations" and its reasoning ability is limited compared to that of humans.

The system excels at analyzing and summarizing complex texts and answering detailed questions, but it does not do as well with future predictions, and it also lacks the ability to recall personal or specific information from one conversation to another, as each interaction is independent.

One of the new features is the ability to describe images and answer questions about them, which is useful for visually impaired people, although OpenAi has delayed the release of this feature for security reasons.

GPT-4 can handle text as long as 25,000 words, making it easier to create extended content and conduct in-depth conversations.

However, as with other similar technologies, it can be induced to unexpected and disturbing behavior.

Although it can pass complex exams and provide medical advice, it can fail at simple tasks such as basic arithmetic.

Despite its advanced capabilities, GPT-4 cannot match human intelligence, especially when it comes to understanding humor or making new hypotheses.

The model was trained with significantly more data than the previous GPT-3 model,5 but OpenAI has not disclosed exactly how much data was used.

In the future, OpenAI plans to develop systems that can handle sound, video, as well as text and images, while continuing to improve the capabilities and accessibility of the technology.