ChatGPT updates: OpenAI held Apple-style presentation
OpenAI announced new chatbot updates (Getty Images)
OpenAI held its first developer conference, known as DevDay 2023, where the creator of ChatGPT and other artificial intelligence products presented numerous new features and capabilities, according to the OpenAI company's website.
GPTs
One of the significant new features is GPTs, custom versions of a chatbot that even users with no programming skills can create to address specific tasks. Users can customize their versions of ChatGPT in a dialogue format by providing instructions and additional information that the algorithm will use to generate responses.
Compared to the Custom Instructions feature available in ChatGPT, GPTs can make multiple custom versions of the chatbot for different tasks.
Subscribers to the Plus and Enterprise plans will gain access to GPTs this week, and soon it will become available to other users as well. By the end of November, the GPT Store is planned to be launched, where users can share their versions of the assistant. Additionally, OpenAI promises to support creators of the most popular GPT configurations by sharing profits with them.
GPT Store (photo: OpenAI)
GPT-4 Turbo
An important upgrade is the GPT-4 Turbo model, which is now used in the paid version of ChatGPT. It is based on data as of April 2023, which is a significant improvement compared to the previous version, which had information only up to September 2021.
The size of the context has also been increased, allowing ChatGPT to process larger volumes of information. The old free version, GPT-3.5 Turbo, can now handle a context of 16,000 tokens.
The performance of GPT-4 Turbo has been optimized, reducing the cost of the API. Now, processing one token at the input costs $0.01 (compared to $0.03), and at the output, it costs $0.03 (compared to $0.06). The API cost for previous GPT versions has also decreased. The ChatGPT Plus subscription remains available for $20 per month.
Additionally, ChatGPT Plus subscribers no longer need to switch between modes for generating images through DALL-E, internet searching, and data analysis, as the algorithm can autonomously determine which function to activate depending on the task.
