Presentations, posters and visuals

The deck is not the talk

The most common failure in student presentations has nothing to do with AI: slides full of text, read aloud. A model will produce that instantly and in volume, so the failure is now available faster.

What is assessed in a presentation is almost never the slides. It is whether you can explain something, in order, to people, and answer questions about it. The slides support that. If your slides could be read without you, you have written a document and are now reading it aloud, which is the least useful form either could take.

Making slides that work

• One idea per slide. If it needs two, it needs two slides.
• The heading is the point, not the topic. Not Results, but Response rate doubled in the treatment group. A reader skimming your headings should get the argument.
• Six words on a slide is plenty. The words in your mouth are the content.
• A figure beats a bullet list, if the figure shows something.
• Anything you would read aloud does not go on the slide.

A legitimate and useful AI request:

Here is my talk, written out. Suggest where a slide should change and what single point each slide should carry. Do not write slide text.

The talk exists first. The slides follow from it. Working the other way (slides first, then working out what to say) is how you end up reading.

Free tools, named against the paid ones

• Slides: PowerPoint and Keynote → LibreOffice Impress, Google Slides, or reveal.js if you want them in a browser and in version control.
• Diagrams: Visio, Lucidchart → draw.io (free, browser or desktop), Inkscape for anything that needs to be publication quality, Mermaid for diagrams written as text.
• Charts: Excel → LibreOffice Calc, Google Sheets, or matplotlib in Python for anything you will need to regenerate.
• Posters: InDesign, Illustrator → Inkscape or Scribus, both free and both entirely capable of an A0 conference poster.
• Images: Photoshop → Photopea in a browser, GIMP, Krita.
• Video: Premiere → DaVinci Resolve free edition, Shotcut, Kdenlive.
• Audio: → Audacity.

Every one of these runs on a modest machine, and several run in a phone browser.

Generated images in academic work

A distinction worth making clearly.

Decorative images (a background, an illustration on a title slide) are usually fine, and you should say they were generated if your institution asks.

Images that carry information (a diagram of a mechanism, a chart, a figure) must not be generated. A generated diagram is a picture of what a diagram looks like, and it will contain plausible wrong labels, arrows in the wrong direction, and structures that do not exist. Draw it in Inkscape or draw.io, or plot it from data.

Anything representing your data must come from your data. Generating or beautifying a figure that represents results is research misconduct, not a style choice, and the line is not blurry.

And copyright here is unsettled. Whether generated images are protected, who owns them, and whether their production infringed the works they were trained on are live legal questions with different answers in different countries and pending litigation in several. If you are producing something that will be published rather than submitted, that uncertainty is a reason for caution. This is not legal advice, and anyone who tells you the position is settled is ahead of the courts.

Rehearsal, which is the actual preparation

Slides get all the attention and rehearsal decides the outcome.

• Say it out loud, standing, at least three times. Silently reading through is not rehearsal.
• Time it. Almost everyone overruns, and being cut off mid-argument is the worst possible ending.
• Rehearse the transitions between slides specifically. That is where people stumble.
• Prepare for questions, which is where a model is excellent: paste your talk and ask for the ten hardest questions an examiner would ask. Then answer them out loud. This is the single best use of these tools for a presentation, and it is exactly the viva preparation from Module 8.

Posters

Same principles, one addition: a poster is read standing up, at a distance, by somebody deciding in four seconds whether to stop. Title readable from three metres, one clear finding, and the rest is a conversation starter rather than a paper in small type. Scribus and Inkscape do this free, and your department’s plotter usually prints for nothing or close to it.