According to OpenAI’s definition, artificial general intelligence (AGI) is a highly autonomous system that outperforms humans in most economically valuable tasks. Although GPT-5 has greatly improved programming and creativity, CEO Altman believes it still falls short of AGI standards.
“I think the way most of us define AGI, we're still missing something pretty important, a lot of things pretty important. It's not the model that continuously learns as it deploys from what it finds, which I think should be part of AGI,” he said.
Altman likens the new version to “a PhD-level expert in your pocket,” whereas the previous ChatGPT was equivalent to a college student and older versions were more like high school students.
Many outstanding improvements
OpenAI says GPT-5 makes fewer factual or “illusory” errors, is better at programming software, and can create full websites and apps. It also has a boost in creative writing, and the model is “less sycophantic” than before, a point the company has emphasized after admitting that ChatGPT was too pleasant, and sometimes annoying, for users.
The new version still supports multimedia – text, images, voice – and adds an “agent” feature that allows for complex tasks like finding available restaurants, shopping online, and even accessing the user’s Gmail, Google Calendar, and contacts if allowed.
In the health space, GPT-5 is trained to better answer medical questions and proactively warn of serious physical or mental risks.
At the launch event, a woman who had been diagnosed with cancer said that ChatGPT helped her think carefully before deciding to undergo radiation therapy. However, OpenAI also affirmed that the chatbot cannot replace the advice of medical professionals.
At the presentation, OpenAI staff demonstrated how GPT-5 could write hundreds of lines of code in seconds and create a complete French-learning program. The technology could also write a more “nuanced” eulogy, demonstrating its ability to handle emotional requests.
GPT-5 “dramatically improves” at avoiding flattery and providing helpful responses within safe limits, rather than outright rejecting sensitive requests, said Nick Turley, product manager for ChatGPT.
The AGI race heats up
The launch of GPT-5 comes as tech giants pour billions into the race to develop AGI. Just days ago, Google’s AI division unveiled a new “ world model,” while Meta CEO Mark Zuckerberg said superintelligence was “on the horizon.”
AI company Anthropic CEO Dario Amodei predicts that the technology could replace half of low-level office jobs within the next five years.
OpenAI says GPT-5 will be available immediately to ChatGPT’s 700 million weekly users. The free version will be limited, while the $200/month Pro plan allows unlimited use.
Along with GPT-5, OpenAI also released two open-source models. While it offers some services for free, the company makes most of its revenue from subscriptions to its most powerful packages and integrating AI into enterprise systems.
According to the latest information, OpenAI is negotiating to sell shares held by employees, with an expected valuation of 500 billion USD, surpassing Elon Musk's SpaceX.
Despite its limitations, GPT-5 is a “major improvement” over previous generations and a clear step forward on the path to AGI, a goal that the tech world is fiercely competing to reach, Altman asserted.
Source: https://baovanhoa.vn/nhip-song-so/buoc-tien-lon-nhung-chua-du-de-thay-the-con-nguoi-159550.html
Comment (0)