Without disclosing its technical architecture or the amount of parameters it employs, OpenAI has released an AI language model called GPT-4. 🧠
A paper, however, has been made public that emphasizes its sophistication and capacity to constrain the words it generates in order to stay to predetermined subjects.
The most notable trait of GPT-4 is its aptitude for performing well on challenging examinations like the AP, SAT, and GRE, scoring higher than GPT-3.5 in a number of areas, particularly arithmetic and developing challenging programs on Leetcode. When it comes to input, GPT-4 can support both text and images, enabling it to annotate images or respond to inquiries about them.
As for adversarial questions aimed at tricking GPT into providing banned information, GPT-4 can still be partially deceived, but it is better than GPT-3.5 or ChatGPT due to its additional training with human guidance on such questions. This training has reduced irrelevant answers to adversarial questions by 82%.
GPT-4 will enable greater input of up to 8,192 tokens and includes a specific API for contexts of up to 32,768 tokens, although with a restricted user base, despite not releasing the full scope of the model.
At the moment, each 1,000 input tokens costs $0.03, and each 1,000 output tokens costs $0.06. The price is much more for the 32K variant, and interested consumers must wait in line to obtain it. Individual customers who subscribe to ChatGPT Plus, however, may use it up to a certain extent for free in the future.
Several companies are now using GPT-4, including Duolingo, which has launched Duolingo Max for chatting with bots in the language being learned, currently available in French and Spanish. Stripe is using it to explain API usage and detect fraud.
The utilization of these AI programs to “write” papers for students may lead to many academic dishonesty cases and result in the expulsion of numerous students. While the act of cheating is easily detectable, students may still be tempted to resort to such measures, despite the serious consequences that come with it. 😂