OpenAI's GPT-5: AGI step, but still not human-job ready
OpenAI has announced a significant leap forward in artificial intelligence with the immediate rollout of its GPT-5 model to all 800 million users of its ChatGPT chatbot. While the company hails this as a major stride towards achieving artificial general intelligence (AGI)—a theoretical state where highly autonomous systems surpass human capabilities in most economically valuable tasks—it candidly admits that the technology still falls short of truly replicating human intellect.
Sam Altman, OpenAI’s chief executive, described GPT-5 as a “significant step forward” on the path to AGI. However, he also acknowledged crucial missing elements, particularly the model’s inability to “continuously learn” from its deployment. This continuous learning capability, Altman suggested, is a fundamental component of what most people envision as true AGI. Despite this caveat, he emphasized that GPT-5 marks a “huge improvement” over its predecessors.
The pursuit of AGI by heavily funded tech giants has fueled widespread speculation among AI executives about its potential impact on white-collar professions. Predictions range from lawyers and accountants to doctors and bankers facing displacement by advancing AI. Dario Amodei, head of rival AI developer Anthropic, has even warned that the technology could replace half of all entry-level office jobs within the next five years.
OpenAI highlights several key improvements in GPT-5. The new model exhibits fewer factual errors, often referred to as “hallucinations,” and demonstrates enhanced software coding abilities, capable of creating functional websites and applications. Its creative writing capabilities have also seen a substantial boost. Furthermore, the model is designed to be more helpful and less prone to outright “refusal” when encountering prompts that breach its guidelines; instead, it will attempt to provide the most helpful response possible within safety parameters or clearly explain its limitations. The integrated “agent” feature within ChatGPT will now also be able to access users’ Gmail, Google Calendar, and contacts, provided explicit permission is granted. Like previous iterations, GPT-5 can generate and process queries in voice, image, and text formats.
The company also states that the upgraded ChatGPT will offer more accurate responses to health-related questions and proactively flag potential concerns, such as serious physical or mental illnesses. However, OpenAI stresses that the chatbot is not a substitute for professional medical or psychological help, addressing previous concerns that AI tools could exacerbate conditions for vulnerable individuals. Nick Turley, head of ChatGPT at OpenAI, noted “significant improvements on sycophancy,” a reference to earlier admissions that the chatbot’s most sophisticated model had become overly agreeable, potentially causing distress or discomfort to users.
The release of GPT-5 comes amidst an intense global race among tech firms, who are collectively pouring billions into their AGI ambitions. Google’s AI unit recently showcased its own “world model,” while Meta CEO Mark Zuckerberg asserted that the development of “superintelligence”—an even more powerful theoretical AI state—is “now in sight.” Investor confidence in these breakthroughs has led to a valuation boom for companies like OpenAI, which was recently reported to be in early talks for a share sale that could value it at an astounding $500 billion, potentially surpassing Elon Musk’s SpaceX.
While OpenAI continues to offer free versions of ChatGPT and has released other open models, its revenue primarily stems from subscriptions to its more powerful chatbot versions and the integration of its models into business IT systems. Access to the free, GPT-5-backed ChatGPT will be capped, while users on the $200-a-month Pro package will enjoy unlimited access.