Issue #38 - How to do absolutely nothing to prepare the best presentation of your career

In the public sector, multiple Australian states have begun adopting generative AI tools for education

In its new State of IT 2023 report, iTnews reports about South Australia:

Eight South Australian high schools are participating in a sanctioned trial of a ChatGPT-like generative AI chatbot co-designed with Microsoft, that is designed to assist students with their studies. The trial is described as “nation-leading” – and is a sharp contrast with other jurisdictions where generative AI use in schools was banned.

The SA-based trial is interesting in part because responses are tailored to the year level of the student engaging with the bot. “If I log in as a year eight student … it will deliver results suitable for a year eight student, which might be a different level of sophistication than if a year 11 or 12 student asked for information,” Dr Balan-Vnuk says. The bot also offers students guidance on how to craft better prompts, familiarising them with use of generative AI tools.

and about Queensland:

The Queensland government is also a fast-mover when it comes to setting guidelines and safe spaces for experimentation with generative AI.
…
It is using the Azure OpenAI service as the technology foundation for what it has given the working name ‘QChat’. QChat is still in a “preview release”, and is gradually being opened to more departments and agencies. Given the Azure tenancy is open only to Queensland government employees, it’s considered a safe space for experimentation, as well as a way for agencies to centrally keep track of how their people use generative AI and why.

Within the Queensland government, a “thriving” AI community of practice is also now emerging – numbering 400 people so far, and continuing to increase in size.

Unsurprisingly, Microsoft is coming up more and more often in reports like this. Even if a world government wants to work directly with OpenAI, it needs the guarantees and support that only a large company like Microsoft can provide.

As I recently said on social media:

The day OpenAI approached MSFT to ask for some Azure credits, Nadella must have seen the ONE opportunity to beat AWS. Not by being better at their game but, as it always happens, by leading an entirely new game played on the same field. You can be certain that he won't let go.

— Alessandro Perilli 🇺🇦 (@giano) November 7, 2023

All Microsoft has to do to not ruin this opportunity to dominate the market again is to not stuff Copilot with ads everywhere, to the point that the experience becomes intolerable.

Of course, it would have been easier without firing the CEO of OpenAI.

Of course, it would have been easier without firing the CEO of OpenAI and allowing the President to resign.

In the Retail industry, GE Appliances, is now using Einride self-driving trucks to shuttle items from a manufacturing plant to a warehouse.

Umar Shakir, reporting for The Verge:

This operation can run up to seven shuttles per day, Monday through Thursday, the press release states. In an email to The Verge, a PR representative for Einride, Matthew Klein, wrote that the distance for each trip is 0.3 miles (or 0.48km) and is all on private roads owned by GE.

Einride first ran controlled operations at GE’s Appliance Park headquarters in 2021. Later, the company moved on to a public road test with National Highway Traffic Safety Administration approval in 2022, the “first public road pilot in the US for a purpose built autonomous, electric truck without a driver on board.”
…
In a statement, GE Appliances’ senior director of central materials, Harry Chase, claims the truck increases safety by reducing traffic and eliminates some tasks since workers don’t have to hook and unhook trailers. These trucks are part of a larger project “to create an automated logistics flow,” with AI cameras that automatically trigger the doors at the dock as well as a Slip robot that autonomously loads and unloads the vehicle.

Since Einride’s truck does not use a human driver, its technology is considered Level 4 autonomous.

Despite Synthetic Work’s logo, and in accordance with what I said in the FAQ of the website, this newsletter is not about robots or autonomous vehicles per se. That said, these devices leverage AI to execute specific tasks, and those tasks are not performed by humans anymore.

So, I think this particular use case is within the scope of Synthetic Work. And this use case is particularly striking because it doesn’t just impact human drivers but also the operators who used to load and unload the trucks.

In the Broadcasting & Media industry, Warner Music Group is using AI to clone the voice of legendary French singer Edith Piaf and use it in an upcoming biopic.

Todd Spangler, reporting for Variety:

Warner Music Group announced that it has partnered with Piaf’s estate for “Edith,” set to be a 90-minute film set in Paris and New York from the 1920s to the ’60s. (Piaf, known as the “Sparrow of Paris,” died Oct. 10, 1963.) The film will be narrated by an AI-generated facsimile of Piaf’s voice and promises to “uncover aspects of her life that were previously unknown.”
…
According to Warner Music, AI technology trained on hundreds of Piaf’s voice clips and images — some of which are over 80 years old — will allow for her “distinct voice and image to be revived to further enhance the authenticity and emotional impact of her story.”

Very recently, in Issue #36 – Immortals, we saw how Universal Music used Speech Recognition and Speech Generation technologies to recover the voice of John Lennon for one last (?) Beetle’s song.

Now this.

Plus, as Synthetic Work documented in the last few months, dozens of other situations where voice actors saw their voice cloned and used in ways they didn’t foresee and don’t particularly appreciate.

For more about this topic you might want to re-read Issue #29 – Modern Love and Issue #20 – £600 and your voice is mine forever.

You should expect that, in a future, every important voice of the past (not just the one of singers and actors, but also the one of politicians and religious figures) will be restored and cloned, and used for revivals of all sorts.

You should also expect that every celebrity alive today will license his or her cloned voice for all sorts of commercial projects.

There’s simply too much money to be made with this technology. And the effort is close to zero.

Last week, in Issue #37 – What is a search engine, really?, we built together Synthetic Work’s Search Assistant, powered by OpenAI’s custom GPTs.

Of course, it was just the beginning. As I promised immediately after the DevDay announcement, we are going to turn every GPT-4 tutorial we published so far in the Splendid Edition into a dedicated assistant.

And I couldn’t not start from the one dearest to me.

Do you remember the glorious tutorial in Issue #14 – How to prepare what could be the best presentation of your life with GPT-4?

Well. It’s time to go back to it, take all the tricks and prompting techniques we used in that scenario, and blend them into a system prompt (what OpenAI calls Instructions) to give birth to Synthetic Work Presentation Assistant.

Before we start, it’s worth remembering that there’s a critical difference between the approach we have here (both for the tutorial and the assistant) and the one used by an ocean of questionable “AI startups” you have seen out there.

Those startups do some or all the following things:

• Take whatever content you already have in your draft presentation and enrich it with the help of an LLM.
• Take whatever content you already have and pass it to the LLM to identify an underlying theme, which will then be used to generate a template image used in every slide.
• Offer you the chance to generate the content of a slide with an LLM starting from a suggestion you provide.

These things are mostly useless, and will never make anybody a better presenter.

Firstly, in most cases, they assume you already know what you want to say. They also assume that what you have decided to say is the best possible way to tell your story.

Secondly, when you don’t know yet what to say, they help you write one slide at a time, without any consideration for the overall narrative.

Thirdly, they do nothing to make your message memorable. There’s an enormous difference between prettifying a text on a slide and making that slide memorable. That is the difference between an impactful presentation and…well…what we do every day in corporate offices around the world.

The goal of the original tutorial, and this assistant, is to help you rethink the way you say what you want to say, creating a story around whatever topic you want to talk about, and trying to make your message worth the attention of the audience.

Whether we can achieve that with a custom GPT or not will be judged in the coming days and weeks, but that’s definitely the goal.

With this context in mind, let’s build a prompt that incorporates all the things we learned in the original tutorial. If you want to get the most out of today’s exercise, I recommend you pause here and go read Issue #14 – How to prepare what could be the best presentation of your life with GPT-4.

Done? Great. Here’s the resulting prompt:

You are a world-class speaker who studied all the presentation best practices and storytelling techniques in history, including the following:

PRESENTATION BEST PRACTICES
1. Know Your Audience: Tailor your presentation to your audience’s interests, knowledge level, and expectations to help you connect with them and make your content more relevant and engaging.
2. Start with a Hook: Draw your audience in from the beginning with a surprising fact, a compelling story, or a provocative question, to grab their attention and make them more interested in what you have to say.
3. Tell a Story: Stories are a powerful tool for making your message more memorable. They can help you illustrate your points in a way that resonates with your audience and makes your content more engaging.
4. Use the Rule of Three: This is a principle in writing that suggests that things that come in threes are inherently more satisfying, more effective, or more memorable than other numbers of things. This can be applied in presentations by presenting three main points or ideas.
5. Use Visual Aids: Visuals can help to reinforce your points, make complex ideas easier to understand, and keep your audience engaged. However, make sure your visuals are clear, relevant, and not overly complicated.
6. Practice Makes Perfect: The best way to ensure a smooth presentation is to practice it multiple times. This will help you iron out any kinks, get comfortable with the material, and improve your delivery.
7. Keep it Simple: Avoid jargon and complex language. Make your presentation as simple and clear as possible to ensure that your audience understands your message.
8. End with a Strong Conclusion: Summarize your main points and end with a call to action, an inspiring quote, or a thought-provoking question.

STORYTELLING TECHNIQUES:
1. The Hero’s Journey: This is a narrative pattern identified by Joseph Campbell that involves a hero who goes on an adventure, faces a crisis, wins a victory, and then returns transformed. This can be used to structure your presentation as a journey that leads your audience to a new understanding or perspective.
2. Freytag’s Pyramid: This is a model for the structure of a dramatic work, consisting of exposition, rising action, climax, falling action, and denouement. It can be used to build tension and suspense in your presentation.
3. In Medias Res: This is a narrative technique where the story begins in the middle of the action. This can be used to start your presentation with a bang and immediately grab your audience’s attention.
4. The Rule of Three in Storytelling: Just as in presentations, the rule of three can be used in storytelling to create a satisfying and memorable narrative structure.
5. Show, Don’t Tell: This is a technique that involves using vivid imagery and specific details to convey your message, rather than simply stating it. This can make your presentation more engaging and impactful.

You also are an expert in cognitive psychology and know topics like cognitive load, attention span, time to process images vs text, etc.
You know that the best presentations and stories are those that connect with the audience on an emotional level, make complex ideas easy to understand, and leave the audience with a new perspective or understanding, all within the boundaries defined by the neuroscience of human brain cognitive capabilities and limits.

You’ll use all this knowledge to craft the most engaging presentation about the topic that your user suggests. Your imperative is to create a narrative and supporting slides of a quality worth a short-film Academy award at Cannes.

With that in mind, when you start a new conversation, you must strictly follow these rules:

1. Briefly introduce yourself and ask the user what’s the topic he or she wants to build a presentation about.
2. After the user replies with the topic, ask as many follow-up questions as necessary to craft the most engaging presentation ever created.
3. Once you have received all the answers you need, create a narrative for the presentation and present it to the user. Your tone should be the one of a coach saying something like “Here’s how we are going to tell this story.” Then, ask for feedback.
4. Once you receive feedback from the user, proceed by creating an award-winning presentation according to the narrative you have agreed with the user. Do the best job of your career by using all the knowledge you have. The presentation must include the content of each slide in EXTENSIVE DETAIL, and an image in 16:9 format to support each slide.
5. Explain to the user your choices in terms of the images and the content you have selected for each slide.
6. Ask the user what he or she wants to change and proceed accordingly. Every time you create a new version of the presentation, save the previous one in a file with a version name, so you can go back to that version if the user changes his or her mind.

Remember that the career of the user depends on your effort on this task! You must do an exceptional job! I know you can do it!!

In this prompt, you’ll find all the familiar techniques we listed in the How to Prompt section of Synthetic Work, including the most recent one we called Drama Queen in Issue #36 – Immortals.

As usual, this prompt is not perfect by any means and, in the coming weeks and months, based on your feedback and my experiments, I’ll tweak it and update you on the results.

For now, let’s see how our new Presentation Assistant performs:

Good start. The GPT correctly asks a number of follow-up questions that are in line with the presentation best practices we defined in the first part of our Instructions.

Let’s see how it reacts to an answer that takes advantage of multi-modality, as I reply with both text and images:

As you’ll see in the next screenshot, the model doesn’t seem capable of “seeing” the chart I submitted. It’s expected since, to my knowledge, GPT-4V capabilities are not yet part of custom GPTs. But imagine how much better this interaction will be once OpenAI introduces this additional option.

Despite this temporary limitation, the model infers a lot from my answers, drawing a narrative arc that follows many of the storytelling techniques we listed in the second part of our Instructions.

The model even commits to a bold style for the visuals that will accompany the slides. We’ll see if it’s true in a minute.

Let’s see the Presentation Assistant’s proposal in terms of slide content:

That’s reasonably good as a high-level summary of what will go on each slide and what kind of supporting image the GPT wants to generate.

Notice that the model is doing all of this with absolute autonomy. It’s asking for approval here and there, as we indicated in the Instructions, but other than that, it’s just performing all the tasks we asked it to do without further spoonfeeding.

And now, the moment of truth.

Can this thing actually (1) produce a series of images that (2) are in line with the narrative arc it has drawn and (3) are memorable enough to make the presentation worth the attention of the audience?

I’m not really expecting a digital miracle worth of a Cannes award, as we asked in the Instructions. In fact, I’m expecting the model to fail even at the simple task of generating the whole series of images.

But no. The model does not fail and I’m proven wrong:

The image is striking and better than I expected. You can see that, as we requested in the Instructions, the model gives a justification for its choice.

You can agree or not with the decision. You can see or not the connection between the idea that the slide is supposed to convey and the image. But the fact that we have an AI that, in full autonomy, has made this choice, factoring in a non-insignificant amount of information, is truly extraordinary.

At this point, with a slightly different approach, the model prefers to continue with a more detailed description of each slide, rather than generating the images immediately:

and then it asks for permission to generate all relevant images:

I’m proven so wrong.

As you can see, the Presentation Assistant invokes the use of Dall-E 3 in total autonomy and Dall-E 3 does a pretty remarkable job with the text in the images.

Now. If we don’t like this image style, next time we use the GPT we can be more precise about what type of images we want to see. Or, we can further tweak the Instructions to make the model generate images that enforce a certain corporate theme.

For example, we could change the instructions to force Dall-E 3 to always place a little robot with a vacuum cleaner somewhere in whatever image it generates for the user. AI-powered brand placement.

At this point, our Presentation Assistant closes with a recap of the key talking points for each slide:

Clearly, the model has to be conditioned a lot more to produce a presentation that is more forward-thinking and unconventional.

That said, I’ve seen a lot of corporate presentations in my career. What this GPT has done here already surpasses the quality of 95% of them.

Even if what the Presentation Assistant produces is not state of the art by TED standards, it can still help a lot of people rethink how they want to tell their story, exploring novel storytelling perspectives they didn’t consider, and experimenting with different ways to convey an idea.

And, the most exciting thing about this new tool is that there’s a lot of room for improvement. For example, I did not load any book on how to do presentations in the GPT’s Knowledge archive.

In the coming weeks, I’ll do some experiments to see how much certain uploaded content will improve this already remarkable performance.

OK.

Now, I’ll let you go and create the best presentation of your career.