ACC 300 · Artificial intelligence · Slides
How generative AI works
The PDF is searchable and suitable for printing.
Contents
Jump to a slide
-
1 Generative AI for accounting work -
2 The model uses two kinds of information -
3 Weights hold what the model learned in training -
4 Context holds the material for this one answer -
5 The model works in tokens -
6 The model ranks many possible next tokens -
7 Context changes the probability of the next token -
8 Probability ≠ truth -
9 The same request can produce different results -
10 Check both runs against the source -
11 Many things change between runs -
12 Reasoning and tools can precede the answer -
13 Large documents enter context in pieces -
14 Retrieval can miss an exception -
15 Language fluency and deterministic calculation are different -
16 The context window is temporary working space -
17 The context window is temporary working space -
18 The mechanism tells you what to check -
19 Use AI as a capable collaborator. Keep the work of proof. -
20 Use AI as a capable
First visit on this browser