Services

How diffusion works

Discuss diffusion 🎲

• does training violate others’ rights? (Vincent 2022)
• who owns the resulting copyright? (Vincent 2022)
• what biases did it learn? (Gault 2023)
• should it be restricted from generating some (sexual or violent) images?
  • how do you specify this?

Transformers 101

Transformer-based AI has been likened to powerful auto-complete and uses:

• vector mapping: convert words into numerical vectors that capture meaning
• positional encoding: capture order of words into vectors
• (self) attention: focus on the strength of words’ connections with other words

Unlike predecessors, transformations can be done in parallel, enabling efficient training on terabytes of data using GPUs.

How transformers work

Ask perplexity.ai

1. about its feelings

   • it doesn’t get sad

2. a koan or lateral thinking puzzle

   • it knows ice cream and shark attacks are a “spurious correlation”

3. a math-related question (e.g., Caesar cipher)

   • “If we apply a Caesar cipher with a shift of 3 to the ciphertext ‘fohrsdwud’, we get the plaintext ‘clementia’.” No, it’s “cleopatra.”

Challenge it

1. with a question to which there is no proper answer

   • I haven’t found one yet

2. to generate a plausible but false answer

   • this suggestion didn’t work

3. to change its mind

   • it corrected itself when queried – not that version number should matter

Use it

• to critique a piece of your writing

  • reasonable suggestions; it didn’t make it worse

• to explain a class concept and generate an exam self-testing question

  • impressive!

Discuss transformers 🎲

• human employment (Heilweil 2023)
• sentience (Lemoine 2022)
• creepiness (Newq 2023)
• learning and plagiarism (Mollick 2023)
• moderation (Gillespie 2020)
  • human trauma and scale
  • bots karma farming, misinformation, propaganda
  • verisimilitude (Reagle 2023)

Videos

• Playlist: Practical AI for Instructors and Students