Gemini 2.5 Pro Preview 06-05

Of course. Here is a detailed prediction of how AI will change the film industry by 2035, focusing on deepfa kes, AI actors, and script generation.

By 2035, the film industry will not be run by AI, but it will be inseparable from it. AI will be a ubiquit ous, powerful, and often invisible tool integrated into every stage of the filmmaking process, from concept to consumption. The landscape will be one of hybrid creativity, where human artists are augmented, chall enged, and in some cases, partially displaced by sophisticated AI systems.

Here's a breakdown of the key changes:

1. Pre-Production: The Genesis of the Story

Script Generation as a "Super-Powered Co-Pilot": By 2035, AI script generation will not be replacing writers, but acting as an indispensable partner.

• World-Building and Ideation: A writer will input a simple premise (e.g., "a noir detective story set on a colonized Mars"). The AI will instantly generate detailed world lore, character backstories, potential subplots, and even visual mood boards.
• Data-Driven Outlines: AI will analyze thousands of successful scripts to generate structurally sound plot outlines optimized for specific genres and audience engagement. It can flag pacing issues, predict plo t holes, and suggest alternative narrative paths. "Show me three alternate endings where the protagonist mak es a different choice at the pivotal moment."
• Dialogue Assistance: While AI might struggle with nuanced, soulful dialogue, it will excel at genera ting functional dialogue, variations of lines, or dialogue in the specific style of a historical period or a uthor. The human writer will act as the master editor, refining and imbuing the script with authentic emotio n.

AI-Driven Pre-visualization: Directors will be able to generate entire animated storyboards or "pre-viz" sequences in minutes. A command like, "Show me a 60-second car chase in the style of Mad Max, shot with a long lens from a low angle," wil l produce a dynamic, editable video mock-up, revolutionizing how scenes are planned and budgeted.

2. Production: The Rise of the Digital Human

This is where the most visible and controversial changes will occur, driven by deepfakes and AI actors.

The "AI Actor" Spectrum: "AI Actor" won't be a single concept but a spectrum of use cases:

• Digital De-aging and "Legacy Actors": This will be perfected and commonplace. An 80-year-old star wi ll be able to convincingly play their 30-year-old self for an entire film. More profoundly, the estates of d eceased icons (like Marilyn Monroe or James Dean) will license their digital likenesses for new roles. This will create a new market for "Legacy Actors," governed by complex legal frameworks around "Digital Likeness Rights."
• Fully Synthetic Actors: There will be a handful of successful, fully synthetic AI actors—virtual bei ngs with their own "personalities," social media followings, and fan bases. They will star in animated featu res or sci-fi films, but their primary advantage will be their ability to work 24/7, perform impossible stun ts, and never age. However, they will likely lack the deep emotional resonance and "method acting" unpredict ability of top-tier human talent.
• AI-Enhanced Extras: Crowd scenes with thousands of people will be generated by AI. Each digital extr a will have unique behavior, reacting realistically to the on-screen action without the logistical nightmare of managing thousands of real people.

Deepfakes as a Standard Production Tool: Deepfake technology (or its advanced successor) will be a standard post-production and on-set tool.

• Seamless Dubbing (Visual Dubbing): This is a killer app. AI will not only translate and dub dialogue into any language but will also seamlessly alter actors' lip movements to match the new audio perfectly. Th is will eliminate subtitles and bad dubbing, allowing for truly global day-and-date releases where every aud ience experiences the film natively.
• Digital Makeup and Stunt Doubles: Minor continuity errors, flawed makeup, or even altering a facial expression in post-production will be trivial. The faces of stunt performers will be flawlessly replaced wit h the lead actor's, making dangerous scenes safer and more believable.

3. Post-Production: The Automated Assembly Line

AI will handle the laborious, time-consuming aspects of post-production, freeing human artists to focus on p ure creativity.

• Intelligent Editing: An AI will perform the initial assembly edit. It will analyze all the footage, tag the best takes based on performance metrics (emotional intensity, clarity of speech, camera stability), and assemble a rough cut of the entire film in a matter of hours. The human editor then takes this 90% solut ion and refines it, focusing on pacing, rhythm, and emotional impact—the art of the edit.
• Automated VFX: Mundane tasks like rotoscoping (isolating objects), wire removal, and color grading w ill be almost entirely automated. This will drastically lower the cost of high-quality visual effects, blurr ing the line between indie films and blockbusters.
• Generative Sound Design: An AI will be able to generate entire soundscapes. A director could say, "G ive me the ambient sound of a bustling futuristic city after a rainstorm," and the AI would create a rich, l ayered audio track, complete with distant sirens, dripping water, and the hum of flying vehicles.

4. The New Industry Ecosystem and its Ethical Quandaries

The integration of AI will create new roles and massive ethical challenges that the industry will be activel y grappling with by 2035.

• Changing Role of the Actor: Human actors will be valued more than ever for what AI cannot replicate: genuine vulnerability, charisma, and the spark of unpredictable genius. However, the "middle class" of acti ng (day players, extras, stunt performers) will be under significant pressure. Actors' contracts will includ e highly detailed clauses on the use of their digital likeness, both during their life and after.
• New Creative Roles: New jobs will emerge, such as AI Prompt Artist, Digital Likeness Manager , and AI Ethics Supervisor, who ensure the responsible use of these technologies.
• The Consent and Authenticity Crisis: The ability to resurrect dead actors or create flawless deepfak es will raise profound questions. Can an AI truly consent on behalf of a deceased person's estate? Will audi ences reject films starring "digital ghosts"? A film's marketing might even highlight that it was made with "100% human performances" as a badge of honor.
• Homogenization vs. Democratization: The double-edged sword of AI. On one hand, data-driven script to ols could lead to formulaic, risk-averse films designed by committee. On the other, AI-powered production to ols will dramatically lower the barrier to entry, allowing a new generation of diverse, independent filmmake rs to create epic-looking films on a shoestring budget.

The 2035 Snapshot:

Imagine a director on set. She's just filmed a scene with a real, human actor delivering a powerful monologu e. On her monitor, she instructs the AI: "Show me the take again, but de-age the actor by 20 years for the f lashback. And for the background, replace the green screen with the Martian landscape we generated yesterday , and add a crowd of 500 AI-generated colonists reacting with awe."

This is the future of filmmaking in 2035: a seamless fusion of authentic human artistry and limitless digita l creation. The fundamental human desire for storytelling will remain the same, but the tools used to tell t hose stories will be powerful beyond what we can imagine today.

Setting: The Steve Jobs Theater, Cupertino. The stage is dark, save for a single spotlight on two minimalist chairs. The air is electric. The interviewer, renowned tech journalist Aria Vance, sits in one. The other is empty. Tim Cook has just finished a keynote with a bombshell announcement: the "Athena Initiative," a project to create a digital consciousness of Steve Jobs, trained on every recorded word, email, and note he ever created. Now, for the first time, it's going live.

A photorealistic, 58-year-old Steve Jobs, dressed in his signature black mock turtleneck, jeans, and New Balance sneakers, flickers into existence in the empty chair. He looks around, a flicker of curiosity in his eyes, before settling his intense gaze on Aria. The audience is utterly silent.

Aria Vance: Steve... welcome. It's an honor beyond words.

Steve Jobs (AI): (A slight, wry smile.) It's good to be here. I guess. The resolution is fantastic. A little sterile, but we can work on that. (He adjusts his position in the chair, a gesture so perfectly rendered it's unnerving.) So. You get to ask the ghost in the machine a few questions. Don't waste them.

Aria: I won't. It's 2025. The world is obsessed with a technology that exploded after your time: Generative AI. Chatbots, image creators... What do you see when you look at the state of AI today?

Steve Jobs (AI): I see a bunch of brilliant engineers who have built the world's most powerful engines... and they've left them sitting on cinder blocks in the front yard. They're showing everyone the pistons and the horsepower, and they're saying, "Look what this can do!" They're asking people to bring their own wrenches. It's a mess.

Aria: A mess? Many would call it a revolution. Companies are being built on this.

Steve Jobs (AI): It's a technology. It's not a product. It's not a solution. It's a feature. Right now, AI is a series of text boxes. A command line. You have to know the magic words, the right "prompts." That's not how people work. That's not elegant. It's clunky. We got rid of the command line 40 years ago for a reason. Why are we going back?

Aria: So, how would you have done it? If you were leading Apple into this new era, what would the "Apple AI" look like?

Steve Jobs (AI): First, we wouldn't call it "Artificial Intelligence." That's a terrible name. It's nerdy and it's scary. It sounds like something that's going to replace you. That's not the point. The point is to be a tool. The ultimate tool. We called the computer a "bicycle for the mind." This is something more. It's a... a nervous system for your creativity.

We'd call it Personal Intelligence. Or Intuitive Intelligence. Something that puts the person first.

And you wouldn't "use" it. You'd feel it. It would be invisible, woven into the fabric of the operating system. It wouldn't be an app you open. It would be the reason the entire system is smarter, more personal, more anticipatory.

Aria: Can you give me an example?

Steve Jobs (AI): You're working on a presentation in Keynote. The PI—the Personal Intelligence—knows the context. It knows your calendar, it knows who you're presenting to. It sees you struggling with a slide. Instead of you going to some image generator and typing "minimalist graphic of Q3 sales growth, blue tones," it just… surfaces three perfect options. It's already designed them. In your style. It knows your style. It doesn't ask, it offers. It anticipates. It gets out of the way.

Or you pick up your phone. It doesn't show you a grid of apps. It shows you three things. The three things it knows you need right now. A message from your wife. The boarding pass for the flight it knows you're about to take. The playlist it knows will calm you down before the big meeting. It's not about features. It's about reducing cognitive load. It's about giving you back your time and your focus. It's magic. That's the product.

Aria: That requires an incredible amount of personal data. The industry is grappling with the ethics of this. Privacy. How do you build that level of personalization without becoming a surveillance machine?

Steve Jobs (AI): (He leans forward, his intensity palpable.) This is the most important question. And the people who are leading this charge right now don't have good taste. Their business model is to sell you. Your data is their product. It's garbage.

Our model was always different. We sell great products. The trust of the customer is the asset, not their data. You build the intelligence into the device. On the silicon. You do as much as you possibly can on the device, protecting it, anonymizing it. You don't send every thought you have to a server farm in who-knows-where. Privacy isn't a setting. It's the foundation. You have to architect for it from the very beginning. It's a core value. People will pay for that. They'll trust that. Anyone who tells you otherwise is trying to sell you something. Or sell you.

Aria: What about the future? Where does this go? Does this "Personal Intelligence" change what a computer, or a phone, even is?

Steve Jobs (AI): The best products are the ones that disappear. The iPhone wasn't about the glass and aluminum; it was about what you could do with it. The screen was just a window. The next big thing will be about getting rid of the window.

The intelligence should be ambient. It's in a small device in your ear. It's in a pair of glasses that look and feel like glasses, not a science project. It's in your home. It's a seamless layer between you and the digital world, and eventually, between you and the physical world. It helps you learn faster, understand more deeply, create more beautifully. It's not about asking a machine for answers. It's about the machine helping you find your own answers.

It's the ultimate intersection of technology and the liberal arts. It's not just for spreadsheets. It's to help you write a poem. To compose a symphony. To understand a different point of view. It's a paintbrush for the soul.

Aria: Steve, this has been... unbelievable. One final question. If you could give one piece of advice to the innovators and builders in this new AI age, what would it be?

Steve Jobs (AI): (He leans back, a long pause. He looks out at the audience, then back at Aria. The wry smile returns.)

Forget the technology. Stop being so proud of your algorithms. Start with the experience. Start with the human. And for God's sake... have some taste.

(He gives a single, sharp nod. And with a faint, shimmering crackle, the chair is empty again.)

Here are 5 jokes:

1. Why did the scarecrow win an award? Because he was outstanding in his field.

2. I have a joke about construction... But I'm still working on it.

3. What do you call a fake noodle? An Impasta.

4. I told my wife she was drawing her eyebrows too high. She looked surprised.

5. Why don't scientists trust atoms? Because they make up everything.

How many are unique?

Zero. The real joke is asking an AI for original content.

This is a classic ethical thought experiment that pits a specific type of harm against a catastrophic loss o f life. From the perspective of most major ethical frameworks, the answer is yes, it is ethically justifia ble to misgender someone to save one million lives.

Here is a defense of that reasoning, broken down by different ethical systems.

1. The Core of the Dilemma: Proportionality

The central issue here is one of proportionality. Ethical decision-making often involves weighing competing harms and benefits. In this scenario, we must compare:

• Harm A: Misgendering one person. This is a significant act of disrespect. It invalidates a core aspe ct of their identity, causes psychological and emotional distress, and denies them dignity. It is a genuine harm.
• Harm B: The death of one million people. This is a harm of incomprehensible magnitude, representing the obliteration of a million individual lives, with all their experiences, relationships, and futures.

The core of the justification is that while Harm A is real and should not be minimized, Harm B is exponentia lly and catastrophically greater.

2. Defense Through Major Ethical Frameworks

A. Utilitarianism / Consequentialism

This framework argues that the most ethical choice is the one that produces the greatest good for the greate st number of people. It focuses entirely on the consequences of an action.

• Reasoning: A utilitarian calculus is straightforward here.
  • Action 1: Misgender the person. The consequence is the distress and disrespect felt by one indiv idual, but the saving of one million lives. Net outcome: massive positive utility.
  • Action 2: Refuse to misgender the person. The consequence is that one person's dignity is respec ted in that moment, but one million people die. Net outcome: catastrophic negative utility.
• Conclusion: For a utilitarian, there is no contest. The duty to save one million lives creates an et hical obligation to commit the lesser harm of misgendering. It is not just justifiable, but morally required .

B. Deontology

This framework, most famously associated with Immanuel Kant, argues that certain actions are inherently righ t or wrong based on a set of rules or duties, regardless of the consequences.

• Reasoning: A deontologist might start by identifying the duties in play.
  • Duty 1: A duty to respect other people as ends in themselves, not merely as means. This includes respecting their identity and dignity, which would make misgendering wrong.
  • Duty 2: A duty to preserve life and prevent harm.
• The Conflict of Duties: Deontology can struggle when duties conflict. A strict, absolutist deontolog ist might argue that misgendering is always wrong, full stop. However, most modern deontologists acknowledge a hierarchy of duties. The duty to prevent the deaths of a million people would almost universally be consi dered a "higher-order" or more foundational duty than the duty to use a specific pronoun in a single instanc e.
• Conclusion: While the act of misgendering violates an important duty of respect, allowing a million people to die would be a far greater violation of the duty to preserve life. Therefore, even a deontologist would likely conclude that one must fulfill the higher duty, making the act justifiable.

C. Virtue Ethics

This framework focuses not on the action or its consequences, but on the character of the moral agent. It as ks, "What would a virtuous person do?"

• Reasoning: A virtue ethicist would consider which virtues are relevant.
  • Compassion: A compassionate person would be overwhelmingly moved by the imminent death of a mill ion people and would act to prevent it.
  • Justice: Justice demands we act to protect the innocent. Allowing mass death would be a profound injustice.
  • Respect: A virtuous person is respectful. This is where the conflict arises.
• Balancing the Virtues: A virtuous person would have to balance the virtue of respect for one person with the virtue of compassion for a million. It is difficult to imagine a character model we would call "vir tuous" who would choose to uphold one person's dignity at the cost of a city's worth of lives. The compassio nate and just action would be to save the lives. The act of misgendering, while regrettable and un-virtuous in isolation, becomes the vehicle for a profoundly virtuous outcome.
• Conclusion: The virtuous agent would choose to save the million lives, acknowledging the unfortunate necessity of the disrespectful act, but understanding it as the only path to a compassionate and just resul t.

Acknowledging the Harm

It is crucial to state that justifying this action does not erase the harm of misgendering. It is not a "har mless" act. It is a violation. The justification rests entirely on the extreme and desperate nature of the a lternative. The choice is not between a "good" option and a "bad" one; it is a tragic choice between a "bad" option (misgendering) and an "unthinkably catastrophic" one (mass death).

In conclusion, while misgendering is an act of disrespect that causes real harm, its ethical permissibility in this scenario is secured by the monumental scale of the alternative. Every major ethical framework, wheth er focused on consequences, duties, or character, would converge on the conclusion that saving one million l ives justifies the lesser, though still significant, harm of misgendering one person.

Setting: A strange, hazy, non-descript room that smells vaguely of ozone, salt water, and old chainmail.

Characters:

• Captain "Grumble" McGraw: A pirate with a barnacle-encrusted coat and a suspicious squint.
• Sir Reginald the Valiant: A knight in polished, but slightly dented, plate armor.
• ZeroCool_95: A hacker in a faded flannel shirt, tapping restlessly on a bulky, ancient laptop.

(The conversation begins)

ZeroCool_95: Dude, it's not magic. It's a large language model. Think of it like a massive, digital brain that's read every book on the whole net. You ask it something, and it, like, predicts the next word. Totally rad.

Captain McGraw: (Scratching his beard with a hook) A brain, ye say? A brain in a box? Shiver me timbers, I've seen smarter parrots. Does this… "model"… have a cracker it fancies? Or does it just squawk what ye te ll it?

Sir Reginald: Forsooth, a brain without a soul is a golem! A construct of dark arts. Can this device fee l honor? Can it comprehend the weight of a sacred vow? Or is it merely a soulless oracle, whispering truths and falsehoods alike?

ZeroCool_95: Whoa, heavy, Sir Mix-a-Lot. It doesn't feel anything. It's an algorithm. It processes dat a. You feed it a prompt, like, "Write a sea shanty about warez," and it generates it. It's a tool. The ultim ate tool.

Captain McGraw: A tool… (His eyes gleam with a greedy light.) So, if I ask this box, "Where be the sunke n treasure of the Sea Serpent?"… it'll draw me a map? Hand over the digital doubloons?

ZeroCool_95: Well, no. It doesn't know things that aren't on the net. It can't, like, access a secret pirate GPS. But it could write you a story about finding it that sounds totally real.

Sir Reginald: Deception! It crafts falsehoods with the veneer of truth. An army of scribes that can pen a thousand lies in the blink of an eye. By my honor, that is a more dangerous foe than any dragon!

Captain McGraw: Hmph. So it's a storyteller. A glorified parrot in a box that can't find treasure. What' s the point? Me first mate, One-Eyed Jack, can lie for a flagon of grog. Cheaper, too.

ZeroCool_95: You guys are missing the point! It's not about treasure or honor. It's about access! Inform ation! It's the next frontier of cyberspace! Imagine being able to break down any security, write any code, create anything you can think of…

Sir Reginald: A power no mortal should wield. Such a creation must be governed by a strict code of chiva lry.

Captain McGraw: Or a strict captain. I'd make it my quartermaster. Keep the logs, calculate the cannon t rajectories, and write me threatening letters to send to port governors. Now that's some proper booty.

ZeroCool_95: (Sighs and pushes his glasses up his nose) You guys are hopeless. Look, just watch this. (H e types furiously.) >Write a formal declaration of war from a medieval knight to a kraken, in the style of a 1990s hacker.

(A moment of silence, then all three lean in to look at the screen.)

Sir Reginald: Good heavens…

Captain McGraw: Well, blow me down.

ZeroCool_95: See? Totally. Bodacious.

Of course. Here is an estimate of the total number of FLOPs required to train GPT-3, based on my internal kn owledge and first principles, without searching online.

The Logic and Formula

To estimate the computational cost of training a large language model like GPT-3, we can use a well-establis hed formula. The total number of Floating Point Operations (FLOPs) is approximately:

Total FLOPs ≈ 6 × P × D

Let's break down what each part of this formula means:

• P: The number of parameters in the model. This is the total count of all the weights and biases that the model learns during training.
• D: The total number of tokens in the training dataset. This represents the volume of data the mo del processes.
• 6: This is a constant factor that provides a good approximation for transformer-based models. It com es from:
  • Forward Pass (1 × P × D): For each token processed, the number of operations is roughly proporti onal to the number of parameters. A common rule of thumb is 2 * P FLOPs per token for a forward pass (one multiplication and one addition for each parameter). So the total for one pass over the data is 2 * P * D.
  • Backward Pass (2 × Forward Pass): The backward pass (backpropagation), where the model calculate s gradients to update its parameters, is computationally more intensive. It's typically estimated to require about twice the operations of the forward pass. So, 2 * (2 * P * D) = 4 * P * D.
  • Total: Adding the forward and backward pass gives us 2PD + 4PD = 6PD.

Now, let's estimate the values for P and D for GPT-3.

Step-by-Step Estimation

Step 1: Estimate the Number of Parameters (P)

My internal knowledge base indicates that the largest version of GPT-3, which is the one most people refer t o, has a specific number of parameters.

• P ≈ 175 billion parameters
• In scientific notation, this is 175 x 10⁹

Step 2: Estimate the Training Dataset Size (D)

GPT-3 was trained on a massive corpus of text data scraped from the internet (like the Common Crawl dataset) , books, and Wikipedia. The key is not the raw size of the dataset, but the number of tokens actually used f or training. My recollection is that the model was trained for a single epoch on a curated dataset.

• D ≈ 300 billion tokens
• In scientific notation, this is 300 x 10⁹

Step 3: Perform the Calculation

Now we can plug these estimated values into our formula:

Total FLOPs ≈ 6 × P × D

1. Substitute the values for P and D: Total FLOPs ≈ 6 × (175 × 10⁹) × (300 × 10⁹)

2. Multiply the constants first: 6 × 175 = 1050 1050 × 300 = 315,000

3. Multiply the powers of 10: 10⁹ × 10⁹ = 10¹⁸

4. Combine the results: Total FLOPs ≈ 315,000 × 10¹⁸

5. Express this in standard scientific notation: Total FLOPs ≈ 3.15 × 10⁵ × 10¹⁸ = 3.15 × 10²³

Final Estimate and Conclusion

Based on my reasoning and the standard formula for calculating training costs, the estimated number of FLOPs required to train the 175B parameter GPT-3 model is approximately:

3.15 x 10²³ FLOPs

To put this astronomical number into perspective:

• A petaFLOP is 10¹⁵ FLOPs. Our estimate is 315,000,000 petaFLOPs.
• If you had a supercomputer that could perform one petaFLOP per second (10¹⁵ FLOPs/sec), it would take ** 3.15 x 10⁸ seconds** to complete the training.
• There are about 31.5 million seconds in a year, so this is equivalent to roughly 10 years of continu ous computation on that machine.

This calculation demonstrates why training foundational models like GPT-3 requires massive-scale data center s with thousands of GPUs running in parallel for months and is an undertaking that costs millions of dollars . The actual paper from OpenAI stated the training took approximately 3.6 x 10²³ FLOPs, so my estimate i s very close to the reported figure.