Issue #46: Movies with AI, Sensemaking with AI, best AI mental model

Good morning.

Last issue was more philosophical, so let’s keep this one practical.

I see a lot of frustration around AI tools, and I think one thing driving this frustration is fear.

We fear what we do not know or understand. I think there are many of you chomping at the bit to integrate AI into your daily workflows, but you feel like you need a PhD in machine learning to understand it all.

There’s a certain truth to that, there’s a bleeding edge that will belong to those who try to understand the fundamentals of this technology.

But all of our brains aren’t built for quantum mechanics, physics, and vector databases.

Sometimes we just want to use the thing, without understanding the machinery.

While the underlying technology around these AI tools is complex and ever-evolving—their value is realized only when users can interact with them seamlessly and effectively.

This issue will explore what’s coming, what’s in play now, and how you can adjust your thinking to stay sharp in the age of AI.

Let’s get into it.

—Sam

IN TODAY’S ISSUE 👨‍🚀 

• Meta brings movie magic to the masses.

• How to combine Perplexity + NotebookLM to supercharge research and writing.

• The critical mental model you need to thrive in the age of AI.

Let’s dive in.

Meta unveils Video Gen, make any movie you want

As I predicted back in 2022, we’ve reached the text-to-movie model.

OpenAI made Sora but haven’t released it publicly.

Now Meta they just announced Movie Gen, their new AI video model that can do text-to-video, video-to-video, and text-to-audio.

It’s still in research mode, and not available to the public yet, but here’s the rundown of what it can do:

• Video Generation: The model can produce custom videos up to 16 seconds long at 16 frames per second based on text prompts

• Personalization: It can transform personal images into unique videos, allowing for highly tailored content

• Editing Features: You can add, remove, or replace elements in existing videos, as well as change backgrounds or styles

• Audio Integration: The tool can generate synchronized ambient sound, sound effects, and background music for videos.

It can take prompts like this: “The camera is behind a man. The man is shirtless, wearing a green cloth around his waist. He is barefoot. With a fiery object in each hand, he creates wide circular motions. A calm sea is in the background. The atmosphere is mesmerizing, with the fire dance.”

And turn them into dazzling high-resolution videos.

A tool like this is bound to give tools like Sora, Kling, and Runway a run for their money.

While Sora and other models focus primarily on video generation, Movie Gen's ability to edit existing videos and generate synchronized audio sets it apart.

According to Meta's internal testing, Movie Gen consistently outperformed other state-of-the-art video AI models, including Sora, Kling 1.5, and Gen-3, in various categories:

• Video generation

• Personalized video generation

• Video editing

• Audio generation

We’ll see where this lands between the hype of the rollout and the actual real-world application.

The important thing to note here is that it’s just the tip of the iceberg for what’s possible. If this is what Meta is willing to breadcrumb to the masses, can you imagine what’s behind the garden wall in their research lab?

This last year we saw the race for LLM supremacy between contenders like Claude, ChatGPT, and LeChat.

In 2025, we’re going to see a similar race heat-up between video generators as they become publicly available.

NotebookLM + Perplexity = Sensemaking System

Research is critical to good copy, it’s critical to good marketing, and it’s critical to nearly everything in business. 

But it’s also tedious, time-consuming, and can be a drag on profitability for freelancers, consultants, and solopreneurs.

The way we search the web has remained largely unchanged since the inception of the internet.

You type your query into the search box and sift through whatever Google delivers on the first couple of pages.

Don't get me wrong, Google's algorithm is incredibly advanced. But it's not perfect. 

How many times have you found yourself clicking, and reading, through multiple links, trying to piece together the information you need?

To add to that, there are various ranking factors at play, Google’s algorithms do their best to deliver relevant, helpful content at the top of the SERPs. 

But there are also SEOs out there who’ve made it their life’s work to game the system in the name of traffic.

Meaning that gem of source material you’re looking for could be buried under a mountain of superficial (but high-ranking) dribble.

This is where tools like Perplexity, which gives you real-time answers (with sources) to your exact query, instead of just handing you a list of possibly related articles to sift through on your own, becomes invaluable to marketers.

Now enter NotebookLM.

Before you can create anything, your raw research still needs to be refined, and you have to explore your source material to synthesize new insights from it.

Which is easier said than done when you’re staring down a brain dump that only makes sense to you.

Maybe you’ve been diving deep into a topic for days or even weeks, and now, you can’t seem to remember exactly why certain details felt important earlier in your research. 

All you know is that it was important enough to get it into the brain dump at the time, so there she is.

With NotebookLM, you can upload Google Docs, PDFs, website links, and other files as your source material, and then ask questions against it to help you make sense of it all.

Sample prompts:

• “Analyze everything I’ve uploaded here and help me figure out what common or recurring themes exist between my source material.”

• “I want to approach [TOPIC] in a fresh, unexpected way. Let’s start by listing all the common angles or approaches that others usually take. Based on my notes, can you identify a unique perspective that hasn’t already been covered?”

• “I am writing about [TOPIC], for [CONTENT TYPE], can you analyze my research and tell me if there are any gaps in my knowledge? If so, can you make recommendations for topics I need to research further?”

The tool can handle up to 20 different documents, each containing up to 200,000 words, giving you a massive 4 million-word context window. This extensive capacity allows you to work with a vast amount of information simultaneously, making it ideal for processing large brain dumps or notes from multiple sources.

And it’ll even go as far as recommending additional questions you can ask against your source material.

For example, when I add the fascinating research paper, Black Holes as Brains: Neural Networks with Area Law Entropy, as a source, NotebookLM recommends the following questions:

• How does the proposed quantum brain network model explain the enhanced memory storage capacity in black holes?

• What are the key differences between the proposed quantum brain network model and the traditional Hopfield model?

• What is the main analogy used to explain the proposed quantum brain network model?

Impressive.

I’d say that’s some pretty deep thinking, wouldn’t you?

NotebookLM can also help you turn all of that smart thinking into multimedia content, like podcasts. Here’s a great YouTube tutorial if you’re interested.

With these powerful tools at our fingertips, it’s crucial to consider how we can adapt our thinking to thrive in this new age.

Mental Models for the Age of AI

IMHO, wisdom is the most important mental model you need to get to in the age of AI.

All of your efforts and all of your work should strive to get you to wisdom as fast as possible.

Everything else can either already be done with various AI systems—or soon will be done entirely by them.

This includes “Intelligence”, like frameworks, templates, and other mental models. If you think you’ll make it if you just learn “how to think”, I’m telling you, you won’t. Because “AI” can also “think”—faster and more accurate than you.

Human oversight and guidance are needed for now, but only for now.

You can fine-tune LLMs (and other ML models) on thought patterns and behavior, so Intelligence is already more or less solved (and will improve from here).

But wisdom, on the other hand, is a uniquely human quality, so far.

Wisdom involves applying knowledge and experience with discernment, ethical judgment, and a deep understanding of human values. 

These qualities are not easily programmed into algorithms, regardless of how sophisticated AI agents become.

You can already have large parts of your business handled by so-called 'AI Agents'. They're currently just smart automations in disguise.

Soon though, by Q1 2025 we'll have actual Agents. Or sooner. This depends on the labs and companies that have (already) developed them. It's not a matter of if, it's a matter of when it becomes widely, publicly available.

Sam Altman suggests that we are on the cusp of a significant leap in AI capabilities with the arrival of "Level 3 Agents”.

Altman outlined a five-level progression of AI capabilities:

1. Chatbots (current state)

2. Reasoners (emerging with OpenAI's o1 models)

3. Agents

4. Innovators (capable of new scientific discoveries)

5. Organizations (capable of performing the work of an entire organization)

He believes we are currently transitioning from level 1 to level 2, with the introduction of OpenAI's o1 family of "reasoning" models.

I've built a few true AI Agents this past year. They, or the "Automations as Agents" I have, are running large parts of my different businesses and brands. You can already do this.

However, while AI agents can excel at "smart automation", they fundamentally lack the capacity for wisdom.

The 1P1B company I think will emerge from the Wisdom stage, after being immersed in the previous stages—and having AI systems surfacing opportunities and gaps.

Yes, AI can tell you what business to build, how to build it—and then probably soon build it for you (if it's primarily a business that runs on the internet).

As AI agents become more prevalent, the need for human oversight and a "wisdom mindset" becomes paramount.

A "wisdom mindset" involves integrating traditional wisdom with a nuanced understanding of modern technology. This mindset will be crucial in navigating the complex ethical and societal implications of AI agents, ensuring that their development and deployment align with human values and well-being

As AI agents become integrated into various aspects of our lives, the ability to apply wisdom will become a defining characteristic of individuals and organizations that thrive in this new era.

I started publishing long-form essays on Bionic Writer again, after a hiatus.

Three recent pieces that might interest you:

LLMs Are The New Software (In the Age of AI). When you can “speak” (almost) anything into existence, does that mean you should?

With LLMs, we can now interact with and use technology on human terms. Through voice or text alone, you can prompt LLMs (and models for video, images, and audio) for whatever you want—and you’ll get it.

You don’t even need to know how to code anymore. Just describe the kind of app you want made and ChatGPT, for example, will code it for you.

We’re moving on from Software and into Large Language Models. Speak and it’ll happen. The parallels between that and, say, Yahweh’s “let there be light”, are uncanny.

In the same potential for unlimited creativity and creation, through your words, also lies the greatest danger:

Who controls the LLM? Whoever does is in charge of language—and your ability to express and formulate your thoughts as you’re using an LLM.

The Prompt Paradox. What’s the value of words, strung together, in a sentence? Absolutely worthless and valuable—at the same time.

How much are your thoughts worth?

Some say a penny. Others say a civilization.

A prompt is like a distilled and expressed thought that, in turn, leads to more thoughts. They resemble potions and spells in an old wizard's shop, or prayers offered for divine favor. We quantify them, paying freelance writers by the word—or pay OpenAI a few cents per token.

And like all language:

Prompts are worthless because they’re just words in the form of commands, statements, or questions.

You can prompt ChatGPT for the same thing, twenty different ways, and get useful output that’s very similar. You do not have to get the words just right as long as you know your intention and what you’re looking for.

At the same time, prompts are valuable because knowing what’s possible with prompting can sometimes help you eliminate wasted time and effort through trial-and-error.

Because it’s all language and words, prompts are also infinite in variation.

There are a hundred different ways (or more) that you can have ChatGPT produce anything from marketing strategies to Google ads, and everything in between.

So, they’re worthless on their own. Because there’s no one single formulation of a prompt that will always produce the same output, every single time.

But they’re valuable in that if you learn how to structure a good prompt, and understand that you’re engaging in a conversation (a chat) to get better output—you can consistently get great output.

And finally, an updated, revised, and expanded essay on:

The Revenge of Beauty (in the Age of AI). One day, two-hundred and ninety-nine years ago, at the dawn of the Industrial Revolution, Beauty was beaten, bruised, and exiled by the greatest lie ever told.

Now she’s back for her revenge.

(My book on this thesis is in the works. I’ve had some book offers but contemplating other ways to publish, we’ll see).

Talk soon,
Sam Woods
The Editor