Welcome to Verses Over Variables, a newsletter exploring the world of artificial intelligence (AI) and its influence on our society, culture, and our perception of reality.

Some of Our Favorite Tools

Runway ML: AI Video for All

What it is: Runway, at its core, is an AI video generator that has evolved over the past year to include a suite of tools for creating and manipulating video content. From text-to-image generation and generative audio to lip synching and video transcription, Runway offers a comprehensive set of features for creators. While there has been a lot of recent buzz around AI video tools from tech giants like Google (Veo), OpenAI (Sora), Luma (Dream Machine), and China’s Kuaishou (Kling AI), most of these models aren't yet available to the public. Runway, on the other hand, has been making waves in the creative world for some time now, even being used in the production of the hit film Everything Everywhere All at Once and an A$AP Rocky music video. The platform also sponsors an annual film festival showcasing its users' creations.

How we use it: One of the standout features of Runway is its customization options, which are particularly helpful for those, like us, who aren't naturally gifted cinematographers. Runway allows users to choose a cinematic style and adjust motion effects when creating a video – whether from a text prompt, an image, or another video. We can move the camera, the subject, or individual items within the frame using the Motion Brush, adding a level of control and finesse to the creative process. However, what really sets Runway apart is its ability to train custom models. By uploading a collection of our own drawings, designs, and inspirations, we were able to train the model on our style and create a world that truly reflects our aesthetic (even before the introduction of style reference). Runway also offers an extensive academy with numerous videos to help users navigate its various tools. While we currently use Runway for short clips (ranging from 4 to 20 seconds) as we continue to master the art of motion, we have a sneaking suspicion that it might take some time before this innovative platform begins to give Hollywood a run for its money.

Suno: Instant Hits, AI-Style

What it is: Suno is an AI-powered music generator that creates realistic songs and vocals in seconds. Available as a standalone app or as a plug-in for Microsoft CoPilot, Suno supports an incredible range of musical styles, allowing users to either upload their own lyrics or have the program generate them automatically. With Suno, you can create a fully realized 2-minute song in just 30 seconds, making it a game-changer for musicians, content creators, and anyone needing quick, high-quality music.

How we use it: Honestly, if you know us, we are more likely to be found listening to a podcast than music, but we can't help but be drawn into the addictive world of Suno. The app's explore function is a particular favorite, allowing us to dive deep into a rabbit hole of musical experimentation. We've found ourselves spending hours playing with different genres, lyrics, and vocal styles, marveling at the sheer variety and quality of the music Suno can generate. Suno’s ability to create incredibly human-like vocals is one of its most-remarkable features, at least to our non-discerning ears. And it does have a unique sense of humor in creating lyrics…sometimes.

We’ll be talking about our favorite tools each week but here is a list of what tools we use most for productivity: ChatGPT 4o (custom GPTs), Midjourney (image creation), Perplexity (for research), Descript (for video, transcripts), Claude (for writing), Adobe (for design), Miro (whiteboarding insights), and Zoom (meeting transcripts, insights, and skip ahead in videos).

Intriguing Stories

Apple enters the Race: Apple held its annual developer conference this Monday and finally announced Apple Intelligence. (If you were stuck watching the livestream, like us, we sat through 50 minutes of underwhelming announcements like a new calculator app for the iPad, until someone finally mentioned AI.) Apple’s AI will be integrated into its new operating systems (Sequoia for Macs and ios18 for phones) set for public release in the fall, but there is another twist. First, Apple is partnering with OpenAI to put ChatGPT into Siri for free, and when you ask Siri a question, you’ll get a pop up asking if you’d like to use ChatGPT or not. Once ChatGPT gives you an answer, you also receive a message reminding you to check its work, as AI can be prone to mistakes.

Second, Apple is integrating its own model to do the “less complicated or less prone to error-type things;” search within your own apps. One demo showed a user asking Siri to search her phone (emails and texts) for when her mother was arriving, and their lunch plans. If we were skeptics, we’d think that Apple was setting up OpenAI to take the fall for any hallucinations or issues with AI search, while keeping its own brand pristine and using its on-device model for the user-friendly (but practical and easy) use cases.

Apple also introduced a Private Cloud for its Apple Intelligence. If your phone can’t handle the AI tasks alone and has to reach to the cloud for more computing power, the data will be encrypted and deleted. Apple’s strict privacy controls are in line with its brand (and its business model of making more money from hardware sales than advertisements) and in stark contrast to its competitors (Google and Meta).

Also, Apple integrated image generation into its model, but before you start planning some wild updates to your avatar, hold your horses. In true Apple fashion, Apple is keeping things tightly controlled: users can only choose from 3 styles — none of which are photographic. Apple is letting us make cartoons but it will be hard to make a deep fake or a malicious image in the form of a sketch with AI baked into the metadata.

Surprisingly, Apple and OpenAI also agreed not to pay each other for this new partnership. OpenAI hopes that getting a taste of ChatGPT on your iPhone will lead to more paid memberships, and Apple will get a cut through its 30% app store fee. OpenAI, however, has nothing to worry about - its annualized revenue more than doubled to $3.4BN in the past six months, the majority of which comes from subscriptions.

Ultimately, we are now setting our reminders (or asking Siri to) for the second week of September, so we can get on the waiting list for a new iPhone.

AI’s Got Character: When it comes to AI models, avoiding harmful behavior is just the bare minimum. Anthropic, the company behind the AI assistant Claude, is taking things a step further by focusing on developing its AI's character. They're training Claude to embody traits we admire in humans - curiosity, open-mindedness, thoughtfulness, and more. The goal? To create an AI that behaves well in a richer, more nuanced way. This "character training" is a core part of Anthropic's alignment process, shaping how Claude navigates the diverse landscape of human views and values. Rather than having the AI simply adopt the perspective of whoever it's talking to, or pretend to be objective and unbiased, they want Claude to be honest about its own views while remaining open to others. It's a tricky balance, but Anthropic believes it's key to creating an AI that can interact with humans in a more authentic and valuable way.

Claude is also the model that Anthropic dived into to understand the “mind of a model” that we mentioned last week, and the model got hung up on the Golden Gate Bridge, after some tweaks. Anthropic has now released a tool allowing some developers to steer their own versions of the model. Dan Shipper, of Every, even created a Katniss Everdeen Claude: we can only imagine the kind of dystopian conversations they must be having. We’re not sure what we’d create, but at the moment we sure are fixated on sushi and french fries.

From Chatbot to MP: In a bid to bring a fresh, high-tech twist to British politics, an AI-powered candidate named "AI Steve" is making waves in the upcoming UK general election. Represented by Sussex businessman Steve Endacott, AI Steve is set to appear on the ballot alongside human candidates vying for a seat in Brighton Pavilion. The idea behind AI Steve is to create a politician who's always available to engage with constituents, listen to their concerns, and incorporate their views into policy decisions. Voters can interact with AI Steve on its website, asking questions and sharing opinions, while a sophisticated language model generates responses based on Endacott's party policies. Endacott, who unsuccessfully ran for local office in 2022, believes AI Steve represents a new era in politics, one where elected officials are more connected to their constituents than ever before. By gathering feedback from "validators" – everyday people who score AI Steve's policies weekly – Endacott aims to create a genuinely representative platform. Personally, we’d love the idea of our US politicians spending more time legislating and less time politicking. Still on the other hand, we are already inundated with politics 24/7 ahead of this November election to wish for digital twins. We’ll stick with Ethan Mollick’s interpretation of AI Steve - models are “infinite copies of some guy named Steve, a grad student who is great at coding & art…but makes stuff up when he is pressed.”

The AI Energy Drain: As Google rolls out its shiny new "AI Overviews" feature to a billion users worldwide, it's not just the occasional cheesy pizza recommendation or blinker fluid gaffe that's raising eyebrows. While we're all enjoying the novelty of having our burning questions answered directly (and often incorrectly) at the top of our search results, there's a hidden cost to this convenience: the massive energy drain of generative AI. Researchers estimate that these AI-powered answers guzzle up to 30 times more energy than your run-of-the-mill keyword search. Every day, a large language model like Google's Gemini can spew out greenhouse gases equivalent to a 49-mile jaunt in your gas guzzler. And let's not even get started on the virtual water cooler gossip – just 10 to 50 responses from ChatGPT can evaporate a whole bottle's worth of H2O. To keep up with the insatiable appetite of generative AI, companies are pouring billions into new data centers, which are projected to double their energy usage by 2026. And while many tout their commitment to renewable energy, the reality is that fossil fuels are often the go-to for keeping these power-hungry servers running 24/7. So as we generate our next epic hip hop poem, it's worth considering the invisible cost of our queries and think twice before hitting enter on that next burning question.

— Lauren Eve Cantor

thanks for reading!

if someone sent this to you or you haven’t done so yet, please sign up below so you never miss an issue.

if you’d like to chat further about opportunities or interest in AI, please feel free to reply.

if you have any feedback or want to engage with any of the topics discussed in Verses Over Variables, please feel free to reply to this email.

we hope to be flowing into your inbox once a week. stay tuned for more!

banner images created with Midjourney.