ChatGPT Evolves Beyond its Past Limitations

ChatGPT played a significant role in the advancement of generative AI technology, with its GPT software supporting various AI-powered applications and services. OpenAI has recently announced promising enhancements for ChatGPT and the GPT APIs, aiming to enhance the practicality of AI.

During the "DevDay" event, OpenAI introduced the GPT-4 Turbo model as an upgraded version of the previously released GPT-4 model. This enhanced model possesses greater capabilities and is equipped with knowledge of global events up until April 2023. Notably, it boasts a 128k context window, allowing it to accommodate the equivalent of over 300 pages of text within a single prompt.

The regular version of GPT-4 is available for ChatGPT Plus subscribers, and Bing Chat uses GPT-4 for conversations. Initially trained on data up to September 2021, ChatGPT and the GPT language model may not answer questions about recent events or facts accurately. While GPT-3.5 is limited to data up to January 2022, GPT-4 has been updated multiple times and is now based on October 2023 data. To address this limitation, the web browsing plugin in ChatGPT allows for search on the web for more recent information, although integrating newer data into the model would be more reliable. Bing Chat also relies on web searches to answer most questions.

According to OpenAI, GPT-4 Turbo excels in tasks that require careful adherence to specific instructions, such as providing answers in specific formats. It also offers JSON mode, ensuring that the output will be valid JSON, which is particularly useful for services using the API or programmers.

The existing GPT-3.5 Turbo model, the only one accessible to free ChatGPT users, is undergoing updates. The new 3.5 Turbo version boasts improved instruction following, JSON mode, and parallel function calling. Internal evaluations indicate a significant 38% enhancement in tasks involving generating JSON, XML, and YAML formats.

As of today, developers can access GPT-4 Turbo, and it will soon be made available to ChatGPT Plus subscribers. The availability timeline for GPT-4 and GPT-4 Turbo for free ChatGPT users remains uncertain. However, GPT-4 Turbo aims to be more cost-effective than the initial GPT-4 release, which is why OpenAI initially limited it to paying customers.

Source: OpenAI

Editor's P/S

As an AI enthusiast, I am thrilled about the continuous advancements in the field, and the recent developments surrounding ChatGPT and the GPT APIs are particularly exciting. The introduction of the GPT-4 Turbo model marks a significant step forward, offering enhanced capabilities and up-to-date knowledge. With its impressive 128k context window, GPT-4 Turbo can process and retain a vast amount of information, enabling it to handle complex tasks and provide comprehensive responses.

The improvements made to GPT-4 Turbo, such as enhanced instruction following, JSON mode, and parallel function calling, further enhance its practicality and versatility. These features make it a powerful tool for developers, allowing them to integrate AI capabilities into their applications and services more efficiently. The potential applications of GPT-4 Turbo are vast, ranging from customer service chatbots to advanced language processing tasks.