Unleashing Artificial Intelligence: OpenAI Opens the Floodgates to ChatGPT-4
The most powerful artificial intelligence (for now) may soon make its way into all our applications and software. Until now, ChatGPT-4 was only accessible to a handful of developers. What impact will this change have on businesses and digital users? Here are some insights.
OpenAI (the brain behind ChatGPT-4) is gearing up to open the floodgates even wider for its most powerful artificial intelligence.
Since March, millions of developers have requested access to the ChatGPT-4 API. For the uninitiated, it becomes possible to embed ChatGPT-4 into any software or application with just a piece of code. This is, of course, subject to internet availability and OpenAI being compensated for each use of its artificial intelligence.
At the end of July 2023, OpenAI teams are planning to expand API access to a larger number of developers. They will also increase the usage limits (in terms of requests per hour sent to ChatGPT) that were initially set due to hardware and economic considerations.
Even though OpenAI has allowed developers to integrate ChatGPT into an application, its teams have been cautious. They have conducted a selection process for candidates and applied usage limits, presumably for several reasons:
- Cost control: The more users there are, the greater the need for hardware resources.
- Limiting the risks of copying or theft of their AI. It is possible to create/train a new AI by querying an existing AI that performs well. Google allegedly initiated its AI, Bard, in this way.
- Preventing questionable uses and potential abuses that have occurred, such as in the case of a mental health support company that used ChatGPT to respond to patients.
Following the relaxation of access conditions and the loosening of set limits, it is highly likely that more AI-based applications will emerge in the coming months.
However, developers and businesses looking to enter this field would do well to anticipate the upcoming European regulations. This includes being transparent, clearly mentioning the use of AI to inform users. There's no need to mask the reliance on a third-party tool.
By the way, the issue of data privacy regarding human-AI interactions should be noted. There is no guarantee that these exchanges won't be reused in AI training to enhance its performance. This potentially poses risks of reusing personal or even intimate data.
In practice, in terms of cost, users will need to pay €0.027 per "chunk" of 750 words sent to the AI (prompts or questions), and €0.055 per chunk of 750 words provided by the AI in its responses.
In more elaborate exchanges (considering history, context, dataset), the cost is €0.055/750 words for queries and €0.11/750 words for responses.
However, the bill can quickly increase depending on the audience of the applications involved.
OpenAI has also indicated that its conversational API is the most popular among API users. This means that most developers integrate ChatGPT as a chatbot-type assistant in their applications. For example, a popular foreign language learning application allows users to ask questions to the AI and receive explanations when the application corrects them and points out mistakes.
Regarding its other, less powerful tools/AIs, OpenAI plans to gradually remove them by January 2024. Therefore, developers are encouraged to use newer, smarter, and more resource-intensive models.
As detailed in a previous article discussing its hardware, ecological, and social impact, ChatGPT is particularly energy-intensive and costly. For example, during the launch of ChatGPT-3.5, the most widely publicized version, the tool cost OpenAI approximately €100,000 per day in infrastructure.
[Cover photo: Mariia Shalabaieva]
Note: The main source of this publication is the OpenAI blog post.
