Version 1.1
When I first heard mention of Chat GPT in tech podcasts that I listen to, I knew it was something special just from the degree of reverence expressed by people who put it to use. I’m a sci-fi lover, so I’ve always been extremely curious about the development of responsive AI that can understand linguistic concepts and reach conclusions through back-and-forth interfacing with users. Given how frustrating my Siri, Google Home and Alexa interactions can sometimes feel, I didn’t think a tool like this was so…imminent.
I’m used to, “Hmmm…couldn’t quite help with that” when I ask for a pudding recipe. I have been following the improvements in AI over the years, so I knew something better was on the horizon. I just didn’t know or expect that the improvement would be as dramatic as this is in a single fell swoop.
ChatGPT seems to almost always have not only a response, but a perfect, thorough and thoughtful response formatted in a way that’s extremely easy to quickly parse. And even if its first response to an inquiry isn’t spot-on every single time, you can usually converse with it to get something useful out of your interaction that sets you on the right path. This kind of tool is game-changing for too many things to cover in a single article, but I want to talk about how I have used this in my workflows as a professional in the post production world, and how tools like this could really change every single industry – dramatically - like, soon!
If there is a problem that I need to solve with a search engine (the one and only: Google), I now stop myself and wonder if ChatGPT would be a better or faster tool for the job? Most of the time, it is. Considering “google” is now a valid verb, according to the dictionary, and I, a 33-year-old, remember being introduced to Google in my 3rd grade computer lab when I was nine-years-old...that is quite an achievement. Nothing has been able to top Google in this arena.
Now, I had heard all this hype about the tool, and I wanted to try putting it to use. Let’s start simple: I found myself in an “it’s on the tip of my tongue” sort of dilemma — a perfect first foray into asking ChatGPT something.
I was working on establishing the folder structure for the post production house, PS260, where I am staffed. I wasn’t quite sure what to call the camera raw/originals subfolder in a way that was both descriptive and concise. I suppose I could just call it “CAMERA-RAWS,” but not all camera originals are technically camera “RAWS” if they are in a non-raw format. I found that label misleading. The dramatically-important question of what to name this folder was killing me because I knew there was something better.
Okay, sometimes you have to coach ChatGPT to understand the tone you’re going for. Instead of having to skim over article summaries, read through a discussion thread to find mentions of this predicament, or worse yet: talking to another human being about the matter, I knew it was the right time to try out ChatGPT. Within seconds, it gave me everything I could have gleaned from a five-minute Google perusal.
(If you’re curious: I landed on “NEGATIVES” – it’s perfect. “Master files” could refer to deliverables, and everything else was either too verbose or not descriptive enough when viewed out of context. “NEGATIVES” feels cool…classic. It pairs well with “DAILIES,” where we stash our proxy transcodes.)
As you can see in how I formatted the inquiry, I stated some context in very-brief, very-plain English, along with my intention. Usually I would literally ask in the form of a question, AskJeeves style, because it’s a language model not a search engine. But ChatGPT is smart enough to understand that I am seeking this information out, even though all I did was state that I am trying to do something. And unlike an asshole on a post production forum or subreddit, I wasn’t met with a sarcastic, “Well good luck with that,” or a sassy “You’re really overthinking this, just call it something, anything, it literally doesn’t matter at all. You have too much free time.”
ChatGPT is simply here to help
It really is a Swiss Army knife when it comes to putting your finger on something that you’re having trouble with. Maybe an agency client is trying to find some underground ‘80s pop hits that are now forgotten and rarely mentioned. ChatGPT will literally know exactly what to suggest based on data, weird niche discussions on random websites, articles, etc., that it already knows and understands. It will instantly provide me with more and sometimes better examples than the hipster-est of music aficionados.
According to my friend, who works in entertainment development, use of neural network AI is already extremely commonplace in the preproduction phase, with mockup art, copy, ideas and brainstorming being either assisted or fully generated by tools like ChatGPT, Dall-E, Midjourney and others. And speaking of art generation, quite a lot can be done in the way of using AI to automatically create visuals based on references, footage, color palettes and other tools to democratize implementations/executions of certain visual styles that would normally be created by a team of rotoscopers, artists and animators. I’m referring in a roundabout way to the exciting-yet-controversial results of the AI-assisted live action to anime conversion, Anime Rock, Paper, Scissors by Corridor Crew on YouTube.
Other great use cases in the world of post production include: finding resource websites and articles; finding a specific piece of art or media that might inspire you related to your description of a mood and purpose; figuring out what to call a certain genre of music based on descriptions or multiple examples (great for finding stock music); or finding interesting ways to rephrase something you’re trying to communicate, like I attempted earlier (it can be hit or miss with things like humor).
Try it yourself: “Tell me how to say ‘goodbye’ to someone in a way that feels harsh or venomous.” The results of that conversation might eventually either inspire or directly provide an advertising copywriter with the perfect catchphrase you need for the closing tagline in your edgy deodorant commercial that you’re workshopping.
Google is still great for things, like literal synonyms and definitions, language translations, etc., because it’s always at your fingertips, but keywords are much less forgiving than being able to actually communicate with your tool like it was a person. It’s extremely intuitive because we have been practicing it since we were literal babies. I’ll still probably use search engines to find many things, but man, I could see that method of finding what you’re looking for becoming akin to the Dewey Decimal system once AI becomes more integrated into our interfaces.
Despite this sudden new power I’ve been introduced to, because I am human, I found myself greedily wishing that it could do more. Perhaps I could ask it something with more specific parameters?
“Can you give me a song that would fit perfectly in an ‘80s movie about people working on Wall Street? Something with no lyrics. Ideally, these would exist on TikTok’s Pre-Cleared Music Library for Ads, or is otherwise licensable on a stock music library.
You’d be surprised. Sometimes it would actually give me the answer to a question quite a lot like that. But in this case, it ended up searching TikTok’s general music library and providing suggestions that were not remotely licensable for small, branded social content. We didn’t exactly have the Super Bowl budget required for “Take on Me,” by A-ha for the spots in question.
So when it failed to get me exactly what I was looking for, I criticized it, right to its “face.” I gave it a “thumbs down,” and explained why it was a failure. It felt very mean. But responding truthfully, with blunt criticism, is definitely encouraged, because ChatGPT learns what it isn’t doing well based on how people interact with it. It responded by confirming that it now understands that there is a difference between TikTok’s general and pre-cleared music libraries, and said it might not be able to help me with something specific in this exact use-case. I’m sure that comes up a lot when you’re asking it to understand something conceptual in a context where your parameters are overly specific. The layers of its algorithms can’t quite parse your criteria and be familiar enough with a specific targeted source — like a specific stock music library — to be able to put its metaphorical finger on what you need.
But that interaction didn’t end there, like it would have with Siri, Google or Alexa (all of whom would not have even had a baseline response). Despite this AI not getting me the exact right answer, it did explain what genres and types of music were commonly used in ‘80s movies about people working on Wall Street, even explaining which ones were especially popular and why. And it told me some keywords that might be associated with those genres so that I might be able to search on specific libraries, like the one I mentioned. (Yes, it offered that solution to me without even being prompted.)
It really did all the work for me, and I had 20 spots to cut so I couldn’t exactly spend time listening to the Other People's Money soundtrack to figure out what sub genre of synth wave is used as transition music between scenes. I was able to find what I needed much more quickly than if I were to have just used my tiny human brain, or spent an hour semi-aimlessly scrolling through stock music for the perfect track. And if I felt like there was more I could get out of it, I would simply ask it to provide more answers, or I would ask follow-up, clarifying questions. When my tired brain is in goblin mode and I can’t even reliably remember what year it is, ChatGPT is there to rescue me. I’ve only been using it for a few weeks and it’s already proven itself to be a major shortcut for technical thinking, creative thinking, communication, information finding and research.
All this new power got me thinking about the future of post production. A computer having this kind of understanding of human language could be extremely powerful (and scary) when integrated with other workflows and software. Imagine if you could descriptively ask it to do something that a series of actions might not be able to predictably or quickly do in Avid, Premiere, Resolve or Final Cut Pro?
“Hey ChatGPT Premiere Plug-in-Edition, after Premiere auto syncs all of my dailies and production audio with its clunky multicam workflow, can you trim all of the audio that precedes or succeeds the longest piece of footage in each respective multicam sequence? Then can you please set a white marker every time you detect the word ‘action’ and ‘cut’ so that I can quickly navigate to usable footage? Then also duplicate those sequences and trim off all the unusable material before the ‘action’ marker and after the ‘cut’ marker, and add ‘_TRIMMED’ at the end of each new sequence’s name? Also, save that series of tasks as a command called ‘breakdown my footage for me’ so you know what I’m asking you to do in the future in less words. Thanks so much, you’re a doll.”
Instead of coding a script to do this series of algorithmic actions (requires knowledge and time) accurately for every little situation like this that an editor can fathom (of which there are infinite), AI could take your human language inquiry and utilize the software’s tools, objects and actions to perform exactly what you are asking of it. It could even ask you clarifying questions to make sure it does everything accurately and effectively without potentially committing unintended actions that have undesired consequences.
With that inevitable function of AI absolutely being on the horizon, let’s just say, as someone who currently earns most of his income through commercial assistant editing, I’m working really hard on trying to become an editor before a ChatGPT successor renders me ‘less useful.’ Frankly, a lot of what we do as assistants is very algorithmic. I have already worked things like macro keyboards, multi-step macros, automation scripts, watch folders and other means of automation into my own workflows. But I have to assume that greater minds than mine are working on implementing neural network-based Artificial Intelligence into productivity software across every industry. It’s essentially just a matter of training it to utilize a piece of software’s API or functions, and learning how we use it in our every day, semi-predictable workflows.
Given the criteria, it might come sooner than we can imagine, which will really shake things up in literally every field. Like, really, really soon.
Now, are ChatGPT and AI potential magic bullets for anything and everything? That’s not a real question. Of course they aren’t, and never will be. As long as society is intact, there will always be ways for humans of any technical, creative or service background to contribute to the workforce. And I certainly don’t think it will completely replace assistant editors anytime soon, certainly not on high-end, big-budget projects. Humans are better at solving crises when things occur that aren’t so predictable. But I definitely think our population of assistant editors will shrink dramatically as these tools become more effective. Many of us will need to fill new, hopefully more creative and less-procedural roles, which is what most of us aspire to anyway.
Already, in certain situations and topics, even when I’m operating at my best, ChatGPT inherently has infinitely more context than I do, because it is literally constantly accessing and learning information by being able to interpret the internet and communicate information to and from me as if the internet itself was a singular brain. If humans ever needed something to speak on our behalf, technology like this is going to make the phonographs we sent into space for aliens to potentially receive look a bit cute.
Right now I am using Chat GPT 4, which, for certain technical tasks, is very much worth the $20/month cost over the free version. The things I am asking it to do, like writing and modifying pages of code, and helping me to write Adobe Premiere software in ExtendScript, are things that used to be extremely tedious and frustrating for me, as I am not a software engineer. But now I have a software engineer angel on my shoulder, advising me every step of the way. Even going from Chat GPT 3.5 to 4 was a massive improvement in its capacity to write error-free code and understand the software that I am writing logically, even when I am communicating very complex interactions to it. The pace at which this is improving is really exciting, and also really scary.
Like any new human technology, some folks will use it for really great things, and some will use it for bad. I’ll admit that I am trying to look on the bright side because the current power this tool has is kind of terrifying. I’ve been having very strange dreams lately that evoke stories like West World, Ex Machina and I, Robot, but I am trying my best to educate myself and stay ahead of it all, because the existence of very powerful AI is inevitable. I know that for a fact now because it has already arrived.
When reflecting on technology and industry, I often think of this story my dad always tells me about a great uncle of ours who owned a company that produced leather straps and harnesses for horses and buggies…Unfortunately, this family lore didn’t end with him adapting and prospering in response to the Industrial Revolution.
Conor O’Sullivan is an Assistant Editor at PS260 (www.ps260.com), which has locations in New York City, Los Angeles and Boston. The studio provides editorial, VFX, design and sound services, and has worked with M&Ms, Grey Goose, AmEx, Draft Kings, New Balance, Burger King and Reebok.