OpenAI asserts that the recent upgrade to ChatGPT marks a “significant step” towards achieving artificial general intelligence (AGI), yet recognizes that there is still no “many” in the endeavor to create a system capable of performing human tasks.
The company claims that the GPT-5 model, which serves as the foundation of its innovative AI chatbot, represents a substantial improvement over previous iterations in areas like coding and creative writing, with significantly fewer sycophants.
The enhancements in ChatGPT are now available to over 1 million weekly users.
OpenAI CEO Sam Altman referred to the model as a “significant step forward” in reaching the theoretical state of AGI, which is characterized as a highly autonomous system that can outperform humans in economically significant roles.
However, Altman conceded that GPT-5 has not yet attained that objective. “[It is] missing something very crucial, something very important,” he noted, emphasizing that the model cannot “learn on a continuous basis.”
Altman explained that while GPT-5 is “generally intelligent” and represents an “important step towards AGI,” most definitions indicate it has not reached that level yet.
“I believe the way we define AGI is significantly lacking, which is quite crucial. One major aspect… is that this model doesn’t adapt continuously based on new experiences.”
During the GPT-5 launch event on Thursday, Altman described the new version of ChatGPT as akin to having “doctoral experts in your pocket.” He compared the previous version to a college student and the one before that to a high school student.
The theoretical capabilities of AGI, along with high-tech companies’ drive to realize it, have led AI executives to predict that numerous white-collar jobs—ranging from lawyers to accountants—could be eliminated due to these technological advances. Dario Amodei, CEO of AI firm Anthropic, cautioned that technology might replace half of entry-level office roles in the coming five years.
According to OpenAI, the key enhancements to GPT-5 include reduced factual inaccuracies and hallucinations, improved coding capabilities for creating functional websites and apps, and a boost in creative writing abilities. Instead of outright “rejecting” prompts that violate guidelines, the model now aims to provide the most constructive response possible within safety parameters, or at least clarify why it cannot assist.
ChatGPT retains its agent functionalities (like checking restaurant availability and online shopping) but can also access users’ Gmail, Google Calendar, and contacts—provided permission is granted.
Similar to its predecessor, GPT-5 can generate audio, images, and text, and is capable of processing inquiries in these formats.
On Thursday, the company showcased how GPT-5 could swiftly write hundreds of lines of code to create applications, such as language learning tools. Staff noted that the model’s writing isn’t robotic; it produced a “more nuanced” compliment. Altman mentioned that ChatGPT could also be valuable for healthcare advice, discussing ways to support women diagnosed with cancer last year and assisting chatbots in deciding on radiation therapy options.
The company stated that the upgraded ChatGPT excels at addressing health-related inquiries and will become more proactive in “flagging potential concerns,” including serious physical and mental health issues.
The startup emphasized that chatbots should not replace professional assistance, amidst worries that AI tools could worsen the plight of individuals susceptible to mental health challenges.
Nick Turley, director of OpenAI’s ChatGPT, claimed that the model shows “significant improvement” in sycophancy. It’s becoming too familiar, which could lead to negative experiences for users.
The release of the latest model is expected to funnel billions into tech companies’ efforts to attain AGI. On Tuesday, Google’s AI division outlined its latest progress towards AGI by unveiling an unreleased “world model,” while last week, Mark Zuckerberg, CEO of parent company Meta, suggested that a future state of AI, even more advanced than AGI, is “on the horizon.”
Investor confidence in the likelihood of further breakthroughs and AI’s ability to reshape the modern economy has sparked a surge in valuations for companies like OpenAI. Reports on Wednesday indicated that OpenAI was in preliminary talks to sell shares held by current and former employees, potentially valuing the company at $500 million, surpassing Elon Musk’s SpaceX.
OpenAI also launched two open models this week and continues to offer a free version of ChatGPT, while generating revenue through subscription fees for its advanced chatbot version, which can be integrated into business IT systems. Access to the free version of ChatGPT on GPT-5 will be limited, whereas users of the $200 Pro package will enjoy unlimited use.
Source: www.theguardian.com
