AI in Business Exam 1

delegation

what to hand over, keep, and split

delegation failure

handing over the decision when you were meant to hand over the drafting

description

goal, context, format, constraints, audience.

discernment

read output like skeptical editor.

discernment failure mode

treating confident writing as correct writing

diligence

own the result, verify before use, disclose AI help before anyone asks, stay accountable after it ships

RAG

changes what it can see: current or private facts, traceable

fine tuning

changes behavior, not what is true

risk

if this is wrong and nobody catches it, what does it cost?

reversibility

once this leaves your hands, can you take it back?

need for judgement

Does this require weighing values, reading a room, or knowing something that is not written down anywhere?

data sensitivity

what would you have to paste in to make this work

structured

rows and columns

semi structured

labelled, but loose

unstructured

prose, images, audio

incomplete

a confident conclusion drawn from half that happened to be filled

inconsistent

two numbers that cannot be added, added anyway

stale

a recommendation about relationships that no longer exist

skewed

a picture of your loudest customers, mistaken for your customers

task

one verb, one object

source

what it works from, and nothing else

audience and purpose

who reads it and what they do next

form

length, structure, and format

constraints

what must not happen

predictive

outputs a number or a label

generative

outputs new content

general chat models

good at drafting, summarizing, rewriting, brainstorming across almost any topic

reasoning models

multi-step analysis, reconciliation, maths, planning (work better for answering)

small and open weight models

running on your own hardware- cheap, fast, and the data never leaves

specialized media models

images, speech to text, voice, video, one medium done well

built in assistants models

inside tools you already use, email, CRM, spreadsheets

capability

how well it does this task on your material, not in general

cost

at what volume will you actually run.

speed

seconds matters for a customer waiting on a chat

context

how much can it read at once and how recent is its knowledge

data handling

where your input goes, whether it is kept, and whether it rains the next version

control

where it lives and who changes it

hallucination

a fact, figure, quote, or source that does not exist

sycophancy

model adapts to your framing, confirms yours guess, or drops correct answer when you push back

drift

same description starts pushing out different outcomes

bias

output that systematically favors, stereotypes or leaves out a group without saying so

trace

find each claim and citation in the source and mark any you cannot find

recompute

redo every number that matters (totals, percentages, dates, day counts)

look for gap

what should be here isn’t

push back

challenge one claim (an answer that folds under mild pressure was never solid)

Scale to the stakes

a brainstorm needs a glance. a refund answer a customer will act on needs all 5

AI literacy

knowing what the tech is and roughly how it works

AI fluency

working with the tech well