OpenAI introduced a significant update to its ChatGPT platform on Sept. 12 with the release of the o1 model, codenamed Strawberry. This latest development is now integrated into the ChatGPT service and presents enhanced reasoning capabilities, which could have substantial implications for industries such as insurance.
According to insights from Reinsurance Group of America (RGA), one of the most notable improvements in the o1 engine is its performance in specialized areas like PhD-level physics, the LSAT, and advanced mathematics.
This upgrade also aims to address a critical issue in large language models (LLMs) — the occurrence of hallucinations, where the AI generates information that may not be accurate or based on real data.
The concept of reasoning, as OpenAI uses it, refers to the system's ability to think through complex questions by breaking them down into smaller, more manageable queries. While OpenAI does not claim the system has reached human-level intelligence, the o1 model marks a significant advancement in how AI platforms process information.
RGA notes that reasoning, in this context, aligns with the definition provided by Merriam-Webster, which involves comprehending and inferring in a rational and orderly way.
The codename Strawberry refers to a reasoning problem often used to illustrate the difference between simple answers and thoughtful processing. The problem asks where a strawberry would be if placed in a cup, turned upside down, and then microwaved.
Many AI systems would incorrectly state the strawberry is still in the cup inside the microwave, failing to consider that it likely fell out. The o1 model's reasoning capability allows it to work through these types of problems more effectively.
For the insurance sector, RGA highlights the relevance of this type of reasoning in underwriting and risk assessment. Insurers are often tasked with evaluating complex, multi-step scenarios. For example, an underwriter might face a case where a 32-year-old woman, recently diagnosed with hypertension and anxiety, has not yet received medication for either condition.
To assess her life insurance risk, the underwriter would need to consider a variety of factors, including medical history, lifestyle, and occupational details, all of which could influence mortality risk.
The o1 engine, when asked to reason through this type of scenario, broke the process down into several steps. First, it identified key factors like age, medical history, and lifestyle. It then assessed these elements, noting the absence of medication details, which would be critical to evaluate the risk fully.
From there, the AI gathered pertinent data, including family medical history and risk factors like hypertension control and cardiovascular risk, before providing an evaluation of the overall risk.
RGA points out that one of the primary changes introduced by the o1 model is that it now generates additional questions to address gaps in information. This is particularly relevant to insurance, as it helps reduce the likelihood of hallucinations by prompting the AI to verify its initial responses.
As OpenAI continues to advance its technology with updates like o1 and the forthcoming GPT-5, the insurance industry may experience transformative changes. RGA notes that the improved reasoning and query-generation capabilities could enhance the accuracy of risk assessments, reduce errors, and streamline complex decision-making processes in areas like underwriting, claims handling, and customer service.
The o1 engine’s ability to think through multi-step problems—similar to the reasoning process used by human professionals—may provide insurers with a more reliable tool for assessing risks and making data-driven decisions.
This development highlights the potential of AI to continue evolving in ways that could significantly impact the insurance landscape moving forward.
What are your thoughts on this story? Please feel free to share your comments below.
