Here are some areas where Blue J may not perform well.
Value judgments
Reason: the system does not understand value-based terms like “most important” or “leading”
Examples: Questions that ask for the “leading cases” in a certain area of tax.
Deep reasoning or interpretive questions
Reason: While the system can summarize and retrieve information, it may struggle with deep analytical tasks that require human judgment and expertise.
Examples: Questions requiring nuanced analysis with unclear positions, strategic advice, or interpretation of complex legal principles in a specific context requiring inference
Creating briefs or opinions requiring human judgment and inference
Reason: While the system can assist in drafting basic documents such as emails or memos, and can convert research conversations into a draft output, they may not be able to handle complex, highly customized legal drafting that requires creativity and deep understanding.
Examples: Drafting bespoke opinions that require a high degree of customization and originality.
Queries of content outside tax
Reason: the system will only base its responses on the tax data it has access to.
Examples: Questions that require integrating knowledge of bankruptcy law.
Hypothetical or predictive scenarios
Reason: Predicting outcomes or hypothetical scenarios requires a combination of current knowledge, experience, and intuition that the system does not possess.
Examples: Predicting the outcome of a pending case or providing strategic advice based on hypothetical tax scenarios.
Here are some use cases that are unlikely to work well:
Adding statements like “as a tax professional”
Contrary to common prompting advice, you don’t need to personify Ask Blue J! Handling this complexity is all built in under the hood and governs how ABJ is able to respond so well
Asking it to “provide citations on every sentence”
Inline citations is a product feature that is now live. Asking ABJ to add them will not get the desired response.
Asking it to process documents copy/pasted in
Dumping in a large document as context in the prompt box and asking it to analyze will not work well, as it lacks clarity and precision in the prompt on what you are actually looking for.