The AI revolution doesn’t end with ChatGPT. On the contrary, this technology was only a prologue to the future. OpenAI this Tuesday introduced the new GPT-4, a natural language model that can interpret images, explain or discuss what is seen in them. It’s also “more reliable, creative and able to handle much more nuanced instructions” with written text than ChatGPT, explained OpenAI, which, together with Microsoft, is making another breakthrough in the AI race when its competition is still trying to follow suit. . Previous model.
The main novelty of GPT-4 is the ability to extract information from images, text or formats that combine images and text. “Across a wide range of domains, such as documents with text and photos, diagrams, or screenshots, GPT-4 shows similar capabilities to text-only input,” boasts OpenAI. However, this ability to analyze images “remains a research development and is not publicly available,” the company warns.
Unlike ChatGPT, GPT-4 will not be open to the general public. Access to it will require a subscription to ChatGPT Plus ($20 per month plus VAT) or a developer access contract, the price of which varies based on the number of requests submitted to the system. The new AI also has limited knowledge of current affairs (it only has data up to September 2021) and will not have machine learning from conversations with users.
OpenAI claims that its new product can also argue and speak much better than ChatGPT. It is based on the evolution of its previous conversational model, GPT-3, adapted for more flexible interaction and “chatting” with people. The company calls it GPT-3.5. “A year ago we trained GPT-3.5 as the first ‘test’ of the system. We discovered and corrected some errors and improved our theoretical foundations. As a result, our GPT-4 exercises were (at least for us!) unprecedentedly stable,” they say.
Can pass any exam
“In ordinary conversation, the difference between GPT-3.5 and GPT-4 can be subtle. The difference comes when the complexity of the task reaches a sufficient threshold,” explains OpenAI. As proof of this, they shared the results of the GPT-4 in a variety of academic and specialized professional exams. According to their data, which this media could not verify, GPT-4 not only can pass all of them, but also that its results are close to the results obtained by the best people.
“For example, pass a mock bar exam with a score that places you in the top 10% of test takers; Instead, the GPT-3.5 score was in the worst 10%,” they clarify: “We do not provide specific training for these exams.”
From OpenAI, they explain that they have also improved the reliability of the system and fixed the bugs that made its predecessor vulnerable to hackers, which made it abuse, hate or disinformation, as shown by elDiario.es. “We spent 6 months making GPT-4 more secure. GPT-4 is 82% less likely to respond to requests for prohibited content and 40% more biased than GPT-3.5 in our internal evaluations.”
All analysts expected GPT-4 to arrive in 2023. Especially after Microsoft turbo-capitalized the company with $10,000 million in exchange for priority access to its technology. The multinational company has declared open war on Google, which could revolutionize digital life by incorporating conversational assistants into every company’s everyday tools. Microsoft has already said that this technology will be available in its Bing search engine, Excel, Word or Windows itself. Google, for its part, has confirmed to the same party that it will do the same with Gmail or Google Docs.
Source: El Diario
