Prompt Injection

A prompting technique in which the user tells the AI what was valuable about its last response, increasing the likelihood the next response follows that same direction.

An AI system that can carry out a multi-step task toward a stated goal with less step-by-step human direction than a standard chat assistant, such as researching a solicitation and flagging gaps.

The policies, roles, and review processes an organization puts in place to make sure AI tools are used safely, ethically, and in line with contract and data requirements.

Computer systems built to perform tasks that normally require human judgment, such as reading documents, recognizing patterns, or drafting text.

A systematic error in an AI model's outputs, often reflecting patterns or imbalances present in its training data.

The sequence of intermediate reasoning steps a model works through on its way to a final answer, similar to a visible train of thought.

The amount of text an AI model can take in and consider at one time when generating a response.

A prompting technique in which the user gives further instructions after an AI response to advance the conversation toward the specific output they actually need.

A type of machine learning that uses layered neural networks to recognize complex patterns in data, underlying most modern language and image models.

Further training a general AI model on a narrower, specific dataset so it performs better on a particular task, such as a company's own proposal voice and formatting.

A large, general-purpose AI model trained on broad data and adapted for many different downstream tasks rather than built for one narrow job.

AI that creates new content (text, images, or other material) rather than only retrieving or scoring existing information.