At the first OpenAI event for programmers, the chief executive of creator ChatGPT, Sam Altman, announced an update to artificial intelligence — now capable of understanding long texts and with access to more recent information.
The model released this Monday (6), in San Francisco (USA), is called GPT-4 Turbo. With the improvements, artificial intelligence will have knowledge of facts up to April 2023.
The tool also gains the ability to understand contextual information from around 128 thousand word fragments — corresponding to a 300-page book.
The new version of ChatGPT also gained the ability to interpret which additional functions to use without needing prior commands. Previously, for example, to search the internet it was necessary to select an option from a menu at the top of the screen. Now the platform identifies the need to search based on textual instructions only.
The same was true for other features focused on programming and data interpretation.
Still at the event, Altman announced that ChatGPT reached an average of 100 million users per week, without needing advertising. “It was all done by word of mouth,” he said. Social networks, for comparison purposes, usually publish this data per day.
Currently, 2 million programmers develop applications using OpenAI technology.
Initially, GPT-4 Turbo is restricted to OpenAI corporate customers. This audience pays the AI startup per amount of text generated and does not have a fixed subscription.
The announcement of GPT-4 Turbo, in addition to the technical advances, will also make the language model cheaper by more than 60%. Until then, GPT-4 cost 20 times more than GPT 3.5.
From this Monday onwards, all OpenAI service options for companies become cheaper.
Another new feature announced this Monday was the possibility of developing specialized platforms called “GPTs” within ChatGPT itself. With this, it will be possible, for example, to train a version of ChatGPT dedicated to answering questions about biology, with less chance of error.
This can be done without the need for programming, just with text commands.
Solutions like this were already being tested in large companies, such as the chat app Discord. The company uses AI developed with OpenAI to identify violent messages and provide appropriate responses to users.
Moderating abusive speech on the internet is one of OpenAI’s efforts to generate money from its artificial intelligence solutions.
Developers will be able to sell GPTs in an OpenAI store, similar to Apple’s App Store and Android’s Play Store.
This Monday, OpenAI platforms also gain the functionality to read texts with six different voices. The demonstration was made in English and there is no information whether the feature will work in other languages.
ChatGPT smartphone applications were already capable of receiving voice commands and interpreting images since previous updates.
In a sign of reinforcing Microsoft’s commitment to OpenAI, the technology giant’s chief executive, Satya Nadella, made a surprise appearance at the conference. The support of the company founded by Bill Gates is crucial in the expensive race to the forefront of artificial intelligence.
The OpenAI team develops language models on Microsoft’s Azure supercomputers. In return, the Windows manufacturer has added the AI startup’s resources to its products.