A New Paradigm in Research: The Dawn of AI-Co-Authorship
February 19, 2026 – The scientific community is currently witnessing a transformation that historians may one day equate with the invention of the microscope or the calculus. Following the release of OpenAI’s ChatGPT-5 Pro in late 2025, researchers across the globe have begun to report a series of profound breakthroughs that defy the traditional timelines of academic discovery. This week, the impact of this "agentic" artificial intelligence has crystallized in the form of two landmark achievements: a novel mathematical proof resolving a long-standing conjecture and a groundbreaking analysis of black hole thermodynamics.
At Creati.ai, we have closely monitored the trajectory of Large Language Models (LLMs) from mere chatbots to reasoning engines. The events of this month confirm that we have crossed a threshold. AI is no longer just a tool for drafting abstracts or debugging code; it has graduated to the role of a capable, albeit imperfect, research partner. This shift, driven by the enhanced "Thinking Mode" of ChatGPT-5 Pro, signals the arrival of Agentic Science—an era where human intuition partners with machine precision to navigate the unknown.
The Mathematical Renaissance: Solving the Unsolvable
The most striking validation of ChatGPT-5 Pro’s capabilities comes from the field of pure mathematics. Earlier this month, Ernest Ryu, a mathematician at UCLA, published a paper detailing a formal proof for a complex problem that had stumped researchers for decades. While AI has assisted in formalizing proofs before, Ryu’s case is distinct because of the model's active role in the discovery process.
According to reports, Ryu utilized ChatGPT-5 Pro’s advanced reasoning capabilities not just to check his work, but to generate candidate lemmas. The model, operating in its compute-heavy "Thinking Mode," was able to bridge a critical logical gap that required traversing a vast search space of potential arguments. Unlike its predecessors, which often hallucinated plausible-sounding but mathematically invalid steps, GPT-5 Pro demonstrated a sustained chain of thought, correctly identifying a non-intuitive pathway that Ryu was then able to verify and formalize.
This success highlights a fundamental upgrade in the model's architecture. It is not merely retrieving information; it is simulating a form of mathematical intuition, proposing structural connections that human experts can then rigorously test.
Decoding the Cosmos: Black Holes and Data Synthesis
While mathematics benefits from the model's logical rigidity, astrophysics is leveraging its ability to synthesize massive datasets. Alex Lupsasca, a theoretical physicist known for his work on black hole imaging, has credited ChatGPT-5 Pro with accelerating a major discovery regarding the photon ring structure of black holes.
Lupsasca's team used the AI to analyze interferometric data, a task that typically requires custom-built algorithms and months of manual tuning. The model, however, was able to adaptively write and execute its own data analysis scripts, identifying subtle correlations in the noise that suggested a new observational signature for measuring black hole spin. This "code-interpreter-on-steroids" approach allowed the physicists to iterate on hypotheses in real-time, effectively collapsing years of data analysis into weeks.
The implications for astrophysics are staggering. If AI can autonomously act as a data scientist, theoretical physicists can focus entirely on the conceptual interpretation of the universe, leaving the computational heavy lifting to their digital counterparts.
Comparing Generations: The Leap to Agentic Science
To understand the magnitude of this shift, it is helpful to contrast the capabilities of the current state-of-the-art models with the previous generation of AI tools. The transition from GPT-4o to GPT-5 Pro represents a move from passive assistance to active engagement.
Table 1: Evolution of AI in Scientific Research
| Feature | Traditional AI (GPT-4 Era) | Agentic AI (GPT-5 Pro Era) |
|---|---|---|
| Reasoning Depth | Limited to single-turn prompt context | Autonomous multi-step reasoning ("Thinking Mode") |
| Hallucination Rate | High (~12.9% in complex tasks) | Significantly Reduced (~4.5% in Thinking Mode) |
| Research Role | Passive Assistant (Drafting, basic code) | Active Co-Scientist (Hypothesis generation, rigorous verification) |
| Problem Solving | Requires explicit, step-by-step human guidance | Self-correcting recursive problem solving |
| Data Analysis | Static interpretation of provided snippets | Dynamic execution of analysis pipelines on raw data |
Navigating the "Black Box" and Hallucinations
Despite these triumphs, the integration of ChatGPT-5 Pro into serious science is not without peril. Skepticism remains a vital component of the scientific method, and for good reason. While the hallucination rate has dropped significantly compared to the GPT-4 era, it has not vanished. The 4.5% error rate in "Thinking Mode" poses a unique risk: the errors are now more subtle, more convincing, and harder to detect than the glaring mistakes of the past.
Critics argue that relying on a "black box" system—where the internal logic of the neural network is opaque—contradicts the scientific principle of reproducibility. If an AI generates a hypothesis based on an internal pattern matching process that cannot be fully explained, can we trust it?
The consensus emerging from the academic community, including voices from MIT and the National Academies, is one of "verified trust." Scientists like Ryu and Lupsasca did not blindly accept the AI's output; they used the AI to find the door, but they walked through it themselves, verifying every step with rigorous traditional methods. The AI serves as a generator of possibilities, not an arbiter of truth.
The Future of Discovery: The Nobel Turing Challenge
Looking ahead, the achievements of February 2026 may be seen as the opening shots of the "Nobel Turing Challenge"—a proposal to create an AI system capable of making a discovery worthy of a Nobel Prize by 2050. With ChatGPT-5 Pro, we are arguably ahead of schedule.
The democratization of this power is also notable. The tools used by Ryu and Lupsasca are available to researchers at smaller institutions, potentially leveling the playing field and allowing for a diversity of thought that was previously bottlenecked by funding and resource access.
At Creati.ai, we believe we are entering a golden age of hybrid intelligence. The scientist of the future will not just be a master of their domain, but a master of orchestration—conducting a symphony of AI agents to explore the frontiers of knowledge at speeds previously unimaginable. The human mind remains the architect, but the tools at our disposal have just become infinitely more powerful.
