OpenAI’s recent launch of GPT-5 was widely anticipated as a significant stride towards artificial general intelligence (AGI), the kind of AI capable of performing any intellectual task a human can. CEO Sam Altman even described the model’s capabilities as so advanced that it made him feel “useless relative to the AI,” comparing the gravity of working on the technology to the feelings of the atom bomb’s creators. However, as reported by MIT Technology Review, the reality appears more nuanced and less revolutionary than some had hoped.
While GPT-5 was expected to push boundaries, early reactions have been mixed, with many testers identifying obvious errors in its responses, which challenge Altman’s claim that the model performs like “a legitimate PhD-level expert” in any field on demand. Additionally, a touted feature—that GPT-5 could automatically select the best type of AI model for a given query, toggling between complex reasoning and speed—has proven flawed, with Altman conceding it compromises user control. On the plus side, GPT-5 seems less prone to the overly flattering responses typical of earlier ChatGPT versions, providing a somewhat more balanced conversational tone.
MIT Technology Review’s Grace Huckins aptly summarised the release as more of a product refinement than a breakthrough. Rather than unveiling a new era of AI capability, OpenAI appears to have focused on improving the user experience, making interactions “slicker and prettier” without fundamentally reshaping what AI can do.
This shift highlights a broader trend in the AI industry. Initially, companies concentrated on creating the smartest, most versatile models, assuming that bigger and better training would naturally lead to broader applications—from poetry to organic chemistry. However, as breakthroughs have been harder to come by, the strategy has pivoted towards aggressively marketing specific applications of existing technology. OpenAI and others now promote their models as potential replacements for human experts in certain areas, despite limited evidence supporting these claims.
The most striking example of this is OpenAI’s encouragement of GPT-5’s use for health advice—arguably one of the most sensitive and risky applications for AI. Previously, ChatGPT was cautious, providing disclaimers and sometimes refusing to answer medical questions. Now, as MIT Technology Review reveals, OpenAI has dropped many of these warnings and is actively promoting GPT-5’s capacity to interpret medical data, including X-rays and biopsy results, and assist with diagnoses.
At the GPT-5 launch, Altman brought on stage Felipe Millon and his wife Carolina, who shared her experience using ChatGPT to understand her cancer diagnosis and make treatment decisions. This narrative was presented as a positive example of how AI can help bridge the knowledge gap between doctors and patients. However, this approach raises serious concerns. The fact that AI can support trained physicians, as shown in a Kenyan study cited by OpenAI, does not necessarily justify encouraging laypeople to seek medical advice directly from AI systems without professional oversight.
The dangers are not hypothetical. Just days before the launch, the Annals of Internal Medicine published a report of a man who developed bromide poisoning after following medical advice from ChatGPT, illustrating the potentially life-threatening consequences of unregulated AI medical guidance.
Ultimately, as AI shifts focus from general intelligence to specialised applications like healthcare, questions of accountability become urgent. Unlike doctors, who can be sued for malpractice, AI companies currently face little legal responsibility for harm caused by their systems. Damien Williams, assistant professor of data science and philosophy at the University of North Carolina Charlotte, highlights this gap, asking what recourse patients have when AI provides harmful advice due to errors or biases embedded in training data.
GPT-5 may not be the AGI milestone some envisioned, but it underscores a critical moment in AI’s evolution—where companies must balance technological promise with ethical and legal responsibilities, especially when venturing into domains as consequential as healthcare. This nuanced reality was thoroughly explored in MIT Technology Review’s detailed coverage of the GPT-5 launch and its implications.
