Talk Tenets
Generated on 2025-05-05
This document combines all individual tenet files for machine learning talks.
Talk Tenet: Audience-Centered Design
Tenet: audience-centered-design
Title: Audience-Centered Design
Description: Every talk should be designed with its specific audience in mind, while recognizing that audiences are rarely homogeneous. Like a well-crafted children’s film that entertains both children and adults simultaneously through different layers of content, talks can operate on multiple levels. Technical concepts can be presented with intuitive analogies that engage newcomers, while subtle technical details satisfy experts. This multi-layered approach ensures that diverse audience members each find value, rather than targeting a single idealized audience level. The goal is not to design separate segments for different audience members, but to craft content that reveals different insights depending on the background knowledge the listener brings.
Quote: “The effectiveness of a talk is measured not by what is said, but by what is understood by each unique listener.”
Examples:
- Using an intuitive analogy that helps newcomers grasp a concept while including technical details that experts will appreciate
- Presenting a visualization that works at multiple levels - simple pattern recognition for non-experts and deeper statistical insights for specialists
- Explaining neural networks with both conceptual metaphors and precise mathematical formulations, allowing audience members to engage at their comfort level
- Incorporating cultural or historical references that add richness for those who recognize them without alienating those who don’t
Counter-examples:
- Using only technical jargon that excludes non-specialists
- Oversimplifying to the point where experts gain no value
- Creating a “split personality” talk with obvious beginner sections and expert sections rather than integrated content
- Treating audience expertise as binary rather than recognizing the spectrum of knowledge present
Conflicts:
- Technical Accuracy: Creating multi-layered content might risk imprecision in metaphors or analogies
- Resolution: Ensure metaphors are carefully chosen to illuminate rather than distort the underlying concepts
- Content Modularity: Multi-layered content may be more complex to separate into reusable components
- Resolution: Design metaphors and examples that can stand alone while also contributing to the deeper narrative
- Time Constraints: Addressing multiple levels might require more time
- Resolution: Focus on elegant explanations that work at multiple levels simultaneously rather than sequential treatments
Version: 1.0 (2025-05-05)
Talk Tenet: Slide Simplicity
Tenet: slide-simplicity
Title: Slide Simplicity
Description: Each slide should contain a limited amount of information to ensure clarity and comprehension. Specifically, slides should have a maximum of three bullet points or two formulae. This constraint prevents cognitive overload, allows the audience to absorb information effectively, and maintains audience engagement. Simplicity on slides shifts the focus to the speaker’s verbal explanation while using visual elements as support rather than the primary information carrier.
Quote: “A slide should be a glance, not a stare.”
Examples:
- A slide with a single powerful image and minimal text
- A technical slide with two formula expressions and appropriate annotation
- A slide with three concise bullet points that elaborate on a key concept
Counter-examples:
- Slides with dense paragraphs of text
- Technical slides with multiple formula expressions stacked together
- Slides with numerous bullet points that overwhelm the viewer
Conflicts:
- Technical Depth: Some complex topics might seem to require more information on a single slide
- Resolution: Break complex topics into multiple slides with logical progression
- Comprehensive Coverage: Desire to cover all aspects of a topic might push toward information-dense slides
- Resolution: Prioritize the most important points for slides and use notes for additional detail
Version: 1.0 (2024-06-15)