AI for Actuarial Exams
AI Tools
Major advancements in generative artificial intelligence (genAI or AI) have allowed for the integration of powerful real-time tools to supplement traditional learning methods.
AI has many advantages over traditional instructors, such as being available 24/7, responding immediately, and explaining concepts clearly.
What is GenAI
Generative artificial intelligence is a broad term that covers a wide range of technologies and use cases. As it relates to The Actuarial Nexus, genAI, or simply AI, specifically refers to large language models (LLMs) that are capable of generating human-like text. These models are trained on vast amounts of data and are able to provide responses that are coherent and contextually relevant.
LLMs work by predicting the next group of characters, also known as a token, in a sequence of text. This allows them to generate text that is similar to the training data they were exposed to. A token is approximately equivalent to 3/4th of a word.
Reasoning models, or models that have to ability to apply Chain-of-Thought Prompting to problems, are becoming increasingly reliable for math. These models are capable of solving calculation problems with a high degree of accuracy.
When prioritizing accuracy for math problems, a reasoning model will perform significantly better than a non-reasoning model. However, the response time will also be slower. For conceptual learning or quick conversations, a non-reasoning model may be more appropriate.
If you are conversing with AI and have a question about a new topic, we recommend clearing the conversation for better results.
Benchmarks by Exam
AI systems occasionally produce incorrect or fabricated statements, although these errors have become far less frequent as model quality continues to improve.
Beginning in early 2025, The Actuarial Nexus initiated a benchmarking study to evaluate AI performance on multiple choice actuarial exams, including Exam P, Exam FM, Exam SRM, and Exam FAM.
We used the following prompt on SOA sample questions:
The possible answer choices are A) [choice A] B) [choice B] C) [choice C] D) [choice D] E) [choice E]
What is the correct answer choice? Provide your letter choice in the format, "The correct answer choice is: [letter]", where [letter] is A, B, C, D, or E. Do not deviate from this format. For example, if you think the correct answer is choice B, you would type, "The correct answer choice is: B". Show your work."
The purpose of the benchmark is to evaluate the AI's ability to solve problems without any additional help outside the question and answer choices (similar to test conditions). We then measured the AI's accuracy, calculated as the percentage of problems answered correctly out of all attempts.
Our research indicates that almost all major models are capable of solving problems with well over 95%+ accuracy on Exam P since early 2025, with many models achieving close to 100% accuracy. We will continue to monitor performance on Exam P, but it doesn't appear necessary to continue benchmarking Exam P as performance has been consistently strong.
For Exam FM and Exam FAM, Gemini 3.0 Pro currently demonstrates noticeably better performance than other models. We recommend using Gemini 3.0 Pro for these exams.
For Exam SRM, several questions include diagrams. We currently have not configured the AI to process images, which is a limitation that may impact accuracy. Many models are capable of processing images; this is a work in progress on our end. For questions without diagrams, performance is high.
In terms of the benefits to your exam prep, AI is primarily used as an interpreter rather than a solutions engine. Since all the questions on the platform have solutions, responses should be more reliable than indicated by the benchmarks, since the solution is provided to the AI as context. GenAI's greatest use is in explaining concepts, providing hints, and elaborating on existing solutions.
Available Tools
The Actuarial Nexus offers several AI tools to help students learn and prepare for actuarial exams:
| Model | Description |
|---|---|
| AI Tutor | AI Tutor is the most common use of AI on the platform. It is available for each question and provides instant feedback and explanations. Under the hood, the conversation always begins with the question, solution, and answer choices by default. This is to ensure that the tutor has the necessary context to provide relevant feedback. It also reduces the frequency of hallucinations. |
| Course | The course material includes AI tools that allow you to ask questions about specific key formulas. Simply press the "Ask AI" button below the formula to get started. Follow-up conversations are available to provide more detailed explanations as needed. |
| Comments | Each question has a forum style discussion page where users can ask questions and discuss the problem. In addition to human responses, AI responses are available to provide instant feedback and answer questions. |
| Flashcards | Each flashcard has an AI assistant that allows you to ask questions about the flashcard. Under the hood, the conversation always begins with the card front and back. This is to ensure that the AI has the necessary context to provide relevant feedback. |
| Written Exams | Written exams are automatically graded by AI. Candidates are assigned a score and provided with feedback on their responses. This feature is available for all written answer exams, including Exam PA, ASTAM, ALTAM, and FSA exams. |
All conversations with AI are private and secure. We do not train on user data. We do not store any user data from conversations with AI without your consent.
Several model providers have openly stated that they do not train their models on user data. Despite these safeguards, we recommend keeping the conversations relevant to exam topics, as we do not have any direct control over the data privacy policies of the model providers.
We encourage students to use AI Tutor as the default method when stuck on a problem. If AI is unable to help, consider employing some of the prompting techniques outlined below, as effective prompts can be saved for future use.
If the AI is still unable to provide a satisfactory response, posting the question in the comments section will notify the instructor for a human response.
Available Models
The Actuarial Nexus utilizes state-of-the-art (SOTA) models from several leading providers. The full list of models can be found here.
Given how quickly the field of AI is evolving, it is likely that new models will be released in the future. The platform is regularly updated to include the latest and most powerful models available.
Settings
AI settings can be modified using the button below.
This button is also found in other relevant parts of the platform. Settings are saved locally in your browser, and will apply to all AI interactions. This means that you may need to update the settings if you switch devices or browsers.
The available options are:
| Setting | Description |
|---|---|
| Model | Select the model you would like to use. |
| Reasoning Claude- models | Claude- models have two modes: reasoning enabled and reasoning disabled. By enabling reasoning, intermediary reasoning tokens are included in the response. This is useful for complex problems, but may result in longer response times. |
| Reasoning Effort | Three reasoning levels are available: low, medium, and high. Medium is the default setting. Low is faster and less powerful, while high is slower and more performative. The reasoning level determines the depth of reasoning. |
Prompting Techniques
Prompt engineering involves applying a series of techniques used to improve the performance of AI through the use of carefully crafted prompts. Learn Prompting is a trusted resource for learning about effective prompting techniques.
Default prompts can be configured, modified, and saved using the button below. This button is also available through the AI Settings icon found throughout the site.
Alternatively, you can visit the Prompt Instructions page to configure and save your own prompts.
Saved prompts can be managed through the "Your Saved Prompts" button.
The Markdown editor can be accessed through the "MD Editor" button, allowing for more powerful responses. The section below provides more information on Markdown and LaTeX formatting.
The bottom of the prompt editor also contains a list of suggested prompts that can be used to improve AI performance. More suggested prompts will be added in the future.
The Future of AI
The field of generative artificial intelligence is rapidly evolving, with new models and technologies being developed at a breakneck pace.
Each new iteration of the base model has also resulted in significant improvements in performance and capabilities. This is further evidenced by the fact that the models are performing better and better on benchmark tests, many of which include reasoning, math, and problem-solving tasks.
From a content perspective, AI is used to assist in writing questions, solutions, course chapters, and even these docs. As context windows increase, and response times decrease, more efficient content creation is possible, allowing us to add questions, solutions, and course chapters at a faster pace.