The debut of Claude Opus 4.5 and Fractal’s Fathom-R1-14B has renewed concerns that AI may begin displacing human talent sooner than expected, as models once criticized for hallucinating now demonstrate professional thinking by passing entry-level engineering and medical exams. But experts note that high test scores do not factor in the human judgment, context and lived experience required in practice.
In a previous conversation with business areaFractal’s Chief AI Research and Platforms Officer Suraj Amonkar noted that the company’s mathematical reasoning model, Fathom-R1-14B, achieved a 100 percent score in the advanced mathematics section of the IIT-JEE. Similarly, Fractal’s Vaidya.ai model achieved a score of 83% in the health reasoning section of the PGNEET, a postgraduate medical exam in India.
“AI models perform well on difficult engineering and medical entrance exams because they are trained on massive amounts of data, including textbooks, academic papers, solved problems, and past exam patterns, which helps them learn underlying concepts, reasoning methods, and problem-solving strategies. They do not “memorize” answers in the human sense, but rather identify patterns and relationships across millions of examples, allowing them to solve complex questions quickly and consistently. They also feature logic, calculations and pattern recognition, all of which are essential for competitive testing,” explains Ganesh Gopalan, the co-founder and CEO of Gnani.ai.
While these models can perform extremely well on structured tests, real-world applications still require human judgment, critical thinking, and context beyond exam-style thinking.
Prof. Raghuram Bharadwaj, Lecturer in the Department of AI and Data Science at IIIT Bangalore (IIIT-B), highlighted that while AI can be very accurate, a mistake made by an AI system can have far more serious consequences than a typical human mistake, especially if the system cannot explain its reasoning or understand the real-world context like a human.
Even as AI becomes more powerful, many areas of medicine and engineering will require human participation, as career decisions in these areas will be shaped not only by knowledge, but also by lived experience, real-world judgment, and an understanding of the human context – qualities that AI does not yet possess.
AI in healthcare
“In medicine, diagnosing a patient often requires interpreting subtle clues, understanding their personal history, and making decisions that balance medical facts with emotional, ethical, and social considerations. Likewise, complex projects in engineering require on-site assessment, coordination between teams, negotiations, and the ability to adapt to unexpected constraints, all of which depend on human judgment. AI can assist in these tasks, but cannot fully replace the personal attention, empathy, and situational awareness that professionals bring. No matter how advanced the AI “A human-in-the-loop will continue to be critical, particularly in high-risk, personalized decisions,” he added.
AI and digital workers
Jaspreet Bindra, the co-founder of AI&Beyond, agreed, adding that while strong AI performance raises fears of workplace displacement, the world is witnessing a shift toward collaboration, with digital workers taking on repetitive, data-intensive tasks while humans assume judgment, insight and responsibility. And while AI’s reasoning continues to advance, it still lacks lived experience and moral foundation, which is why it should be treated as a powerful assistant rather than an autonomous expert.
While it may seem worrying that AI is outperforming humans in exams, it ultimately has benefits. As AI takes over tasks based on memory and routine problem solving, students and professionals can focus more on conceptual understanding, critical thinking and creativity. This also forces educators to develop exams that test logical reasoning rather than memorization. In the long term, AI can improve human learning by taking over the routine and freeing people to focus on insights, innovation and judgment.
Published on November 26, 2025