OpenAI, the owner of ChatGPT, announced this Monday (6) for its two main languages - GPT-3.5 and GPT-4 – and confirmed that it will start protecting customers against lawsuits related to the law.
GPT-4 Turbo TSI
- At its first developer conference, the company also announced the GPT-4 Turbo.
- Currently available as a preview, it has been trained with current data back to April 2023.
- The previous version of GPT-4, released in March, only studied from data until September 2021.
- The new GPT-4 will also accept graphics, text-to-speech, and DALL-E 3 integration, allowing it to perform more complex tasks, OpenAI said.
- OpenAI plans to make the final version of GPT-4 Turbo available next week, but no date has been given yet.
Companies using the GPT-3.5 Turbo model, starting in March, will also see improvements in performance and cost, the company said.
Cheaper for developers
OpenAI confirmed that the new version of GPT-4 will be three times cheaper than previous versions and will “see” more data, with a 128K point window – the “equivalent of many over 300 pages of text in one go”.
In practice, this means that a large language model will better understand the question, in addition to generating more answers. Until now, OpenAI has released two versions of GPT-4: one with the definition of windows 8K and the other with 32K.
Prevention of crime
In addition to the development, OpenAI also confirmed that it will provide legal protection to business users through a program called “Copyright Shield“: “We will now step in and defend our customers if they face criminal charges,” the company said in a statement.
It is worth noting that Google has adopted the same approach. The detective giant said it would take legal action if customers using its AI capabilities were charged with crimes. Microsoft will also provide the same protection to business users of Copilot, its AI-powered assistant.