Slide 0: Writing with AI: Empowering Artists & Researchers with ChatGPT

• Welcome!

Slide 1: Who I Am & How I Use ChatGPT

• Hi, I’m Ana – ML engineer, I also like to read
• I use ChatGPT to:
  • Write tech proposals & project descriptions
  • Motivation letters & outreach emails
  • Remove errors and typos from any text or messages
  • Get feedback on my poetry (mostly to feel better)

Slide 2: What is ChatGPT / an LLM?

• ChatGPT is a Large Language Model (LLM) – it completes text based on patterns in its training data.
• It's like a very powerful autocomplete, not a search engine.
• It does not “think” or “understand” – it predicts the next word.

Slide 3: How Does an LLM Work?

Let’s say you prompt ChatGPT with:

"An 8-year-old says: I hate going to..."

ChatGPT completes it with:

"...school."

Not:

"...a strip bar."

Key idea:
ChatGPT doesn’t know what’s appropriate or true.
It just predicts what’s most likely to come next based on patterns in its training data.

Slide 4: Core Limitations

• Confidently wrong (“hallucinations”)
• Lacks real-world experience or emotional nuance
• Struggles with long-term consistency
• May reflect biases and clichés in training data

Slide 5: Basic Use

• Start simple:

  • “Write a 300-word artist statement about…”
  • “Summarize this grant call for me”

• Use the right model:

  • GPT-4o: good for most purposes

• Context length: GPT-4o can process ~128,000 tokens with Pro. That’s about:

  ~300 pages of text

  ~100,000 words

Slide 6: GPT-4o and GPT-4.5 – What's New?

• GPT-4o (released May 2024):

  • Multimodal: accepts and generates text, images, audio, and video
  • Faster, cheaper, and more responsive
  • Available to free-tier users with rate limits; higher limits with Plus
  • Ideal for creative tasks, image input/output, and real-time AI use

• GPT-4.5 (released February 2025):

  • Improved reasoning and coding accuracy over GPT-4o
  • Fewer hallucinations, more consistent output across tasks
  • Stronger performance in scientific, legal, and analytical use cases
  • Available via ChatGPT Pro and enterprise integrations

The AI model lineup evolves quickly – check for the latest.

Slide 7: Prompt Engineering 101

• Be specific: topic, audience, tone, length
• Use roles: “You are a grant-writing coach…”
• Use structure: bullet points, sections, examples
• Iterate: “Rewrite to sound more poetic…”
• Paste your own text for style mimicry

Slide 8: Examples

• "Rewrite this email as a compelling pitch to a curator"
• "Draft a concept note for a socially engaged media installation on migration"
• "Suggest 3 themes in my project that would appeal to public funders"
• "Turn this messy brainstorm into a 1-page artist statement"

Slide 9: Limitations (Again)

• People may recognize AI-written text:
  • Overly smooth, generic phrasing
  • Repetitive or vague
• Watch for:
  • Bias in tone/language
  • Clichés ("art has always questioned society")
  • Empty inspiration quotes

Slide 10: Common Syntax & Punctuation Patterns

• Em dashes everywhere — often for emphasis or rhythm
• Comma-heavy, especially in lists and long clauses
• Balanced sentence structures with symmetry
• Overuse of transition words (however, thus, while, etc.)
• Rarely uses fragments or abrupt shifts — everything is “clean”
• Tends to write in perfect grammar — but sounds overly smooth or impersonal
• Download the PDF handout

Slide 11: Writing in Your Style

• Paste your past work & ask: “Analyze my writing style”
• Then: “Write about XYZ in the same style”
• Tip: Ask it to describe your tone first, then apply it to new drafts
• Great for: bios, mission statements, funding appeals

Slide 12: Web Search with ChatGPT Pro

• Yes, ChatGPT Pro can access the web (if you ask!)
• Great for:
  • Looking up real stats, policy quotes, orgs
• Tip: Say “Search the web for…” or “Find me current…”

Slide 13: Deep Research Mode (ChatGPT Pro)

• OpenAI’s Deep Research tool lets ChatGPT:
  • Autonomously search the web
  • Read and analyze multiple sources (text, PDFs, images)
  • Produce structured, nuanced research reports
• Designed to mimic the work of a human analyst, but faster
• Can take 5–30 minutes per task; access rolling out to Pro users

Slide 14: What Are Custom GPTs?

• Custom GPTs are tailored versions of ChatGPT with:

  • Special instructions (like personality, goals, tone)
  • Optional uploaded files for added knowledge
  • Access to tools (code interpreter, web, DALL·E, etc.)

• Anyone can create one using the GPT Builder:

  • No coding required — just describe what you want
  • Configure behavior, knowledge, and capabilities
  • Test and refine it in real-time

• Used for:

  • Writing assistants, tutors, life coaches, HR bots
  • Art portfolio reviewers, grant writing partners, UX advisors

Slide 15: Are Custom GPTs Better?

• They can be — if well-designed:

  • Stay in character and stick to specific domains
  • Work faster by skipping broad generalizations
  • Handle niche tasks better than generic ChatGPT

• But:

  • Quality depends on good setup and clear instructions
  • Not smarter than GPT-4o — just more focused
  • Can still hallucinate if not properly grounded

• Examples:

  • "Expert Grant Proposal Writer" (haven't tried myself)

Slide 16: Q&A + Live Prompt?

• Any questions?
• We can also write some prompts together

Slide 17: Links

• Deep research mode