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.

Back to Basics

AI Whispering: What is Prompt Engineering?

Prompt Engineering is an intimidating word for the delicate art of sweet-talking AI into doing your bidding, a skill that's part technical wizardry, part smooth operator, and essential, in our increasingly AI-driven world. Think of it as crafting the perfect pick-up line for a robot. Too vague, and you'll get the AI equivalent of a blank stare. Too specific, and you might as well do the task yourself. The trick is finding that Goldilocks zone where your silicon dance partner twirls exactly as you lead. Don’t worry, the tech giants behind the LLMs have made prompt engineering even easier: first you can always ask the model what is the best way to prompt itself (or another model), and if that doesn’t work, you can always turn to the prompt libraries they have provided, serving as both a playground and a classroom. 

Anthropic's offering, hosted on their documentation website, is a collection of pre-designed prompts optimized for their AI models, with tools for everything from business analytics to creative writing. The library even features a user submission option, making a collaborative hub for the prompt engineering community. Not to be outdone, Google's Generative AI Prompt Gallery, nestled within their Vertex AI documentation, features prompts in multiple languages and covering an impressive array of domains. Whether you're looking to generate code, analyze videos, or even process audio, Google's gallery has a prompt for that.

These libraries are more than just collections of clever phrases. They're a window into the art and science of effective AI communication. By studying these curated prompts, users can learn the subtle nuances that make the difference between an AI fumbling in the dark and one that hits the mark with laser precision. We’d recommend playing around in Anthropic’s or asking ChatGPT how to prompt itself, just to get a feel for the best use of the tools.

But perfecting this art requires more than just a good phrasebook. It demands a deep understanding of how AI models think—or rather, process information. That's where skills in natural language processing come into play, forming the technical backbone of this emerging field. The challenge lies in the nuances. A well-crafted prompt can coax an AI into producing stunningly accurate and creative outputs, while a poorly constructed one might lead to nonsensical or biased results. It's a balancing act between specificity and flexibility, requiring prompt engineers to anticipate how the AI might interpret their instructions. The better the prompt, the less hallucination and frustration. This isn't just about getting chatbots to engage in witty banter. Prompt engineering has serious applications across industries. In healthcare, it could mean the difference between an AI assistant providing general health tips and offering tailored, potentially life-saving advice. In finance, it could be the key to revealing more accurate predictive models.

Perhaps the most intriguing aspect of this field is its potential to democratize AI interaction. As prompt libraries become more sophisticated and user-friendly, they're lowering the barrier to entry for individuals and businesses looking to harness the power of AI. It's no longer just the tech giants who can play in this sandbox—with the right prompts, anyone can become an AI Jedi master.

Tool Update

Cursor: The Coding Companion that has Everyone Buzzing

What it is: Cursor is an intelligent code editor which brings a new level of sophistication to your coding workflow. Imagine having a brilliant collaborator who completes your code and anticipates your next move. Cursor does just that, offering everything from writing entire functions to debugging with the precision of a seasoned pro. Its natural language interface means you can edit code with plain English commands, making complex tasks as simple as chatting with a knowledgeable colleague.

How we use it: We haven’t jumped into Cursor yet (because we didn’t consider ourselves a dev), but if this 8-year old can, we suspect we will be soon. We've been watching the feedback on Cursor’s capabilities, and it's clear this tool is a game-changer for developers of all stripes. Novices find themselves building full-stack applications in record time, while seasoned pros are leveraging its power to streamline complex system development. It's particularly impressive in rapid prototyping scenarios - we've seen users create functional habit tracker apps or landing pages in the time it usually takes to sketch out a rough plan. But perhaps its most charming feature is its ability to generate documentation, turning those tedious tasks into a breeze. Whether you're embarking on a new project or wrangling legacy code, Cursor is an indispensable ally in the art of programming. If you (or your 8-year old) jump into Cursor, let us know what you build.

Midjourney: From Palette to Product: Midjourney, our go-to AI image generator, is now dipping its toes into real nuts and bolts. The company dropped a cryptic announcement that they're "getting into hardware," setting the tech world abuzz with speculation. At the helm of this new venture is Ahmad Abbas, a hardware veteran who cut his teeth on Apple's Vision Pro headset. Now, as Midjourney's Head of Hardware, he's tasked with bringing the company's digital magic into the tangible world. The company is being coy about the form of product, ruling out pendants, but hinting at "multiple efforts in flight." Founder David Holz even joked about making an "orb," channeling his inner Saruman. Will we soon be channeling our inner Gandalf, peering into AI-powered crystal balls? The tech oracles are keeping mum.

We’ll be talking about our favorite tools, but here is a list of the 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

AI Apps Get a Glow Up: The New Cool Kids on the Block: Every week, we spotlight our favorite AI tools, and it's no surprise that many of these trusty standbys continue to hold their ground on the latest a16z Top 100 Gen AI Consumer Apps list. But the AI landscape is nothing if not dynamic, and this new ranking reveals a fascinating shift in the digital terrain. In a testament to the breakneck pace of innovation, nearly 30% of the apps on the list are fresh faces since the last report just six months ago. While our tried-and-true favorites remain, we're seeing an influx of creative powerhouses and plucky new challengers making their mark in the AI consumer space. It's like watching the tech equivalent of a blockbuster season premiere – familiar faces mixed with intriguing newcomers, all vying for our attention in increasingly clever ways. Creative applications are dominating the field, particularly on the web. Over half of the top web-based AI products now focus on content generation or editing across various media types. Music generators are seeing notable growth, with Suno climbing from #36 to #5 in just six months. While ChatGPT maintains its top position on both web and mobile platforms, new competitors are gaining traction. Perplexity has secured the #3 spot on the web list, while Anthropic's Claude has entered the top five. These developments suggest a growing market for specialized AI assistants. ByteDance, TikTok's parent company, is making a significant push into the AI space. The tech giant now occupies six spots across the web and mobile lists with various AI applications, indicating a strategic expansion beyond social media. A new category has emerged in the mobile space: AI-powered aesthetics and dating apps. Products like LooksMax AI, Umax, and RIZZ are gaining users by offering AI-driven attractiveness ratings and dating advice, signaling consumer interest in AI for personal enhancement. Discord remains a key platform for AI product growth, often serving as an early indicator of future success. (Many companies use Discord to test and refine their products before launching standalone applications.) As the AI app ecosystem continues to expand, it's clear that consumers are increasingly embracing these tools for creative, productive, and personal uses. The rapid changes in the rankings highlight the dynamic nature of the AI consumer market and suggest further innovations are likely on the horizon.

AI's Curious Blind Spots: A few weeks ago, we discussed the nature of large language models (LLMs) and their token-based architecture. Now, these digital prodigies are serving up a double scoop of humble pie, stumbling over both language and fractions. Some of the world's most advanced AI systems are flunking tests that most third-graders would ace. GPT-4o and Claude insist there are only two "r"s in "strawberry," while many popular LLMs – including ChatGPT-4o – confidently declare that 9.11 is larger than 9.9.(This mathematical quandary has since, been solved, luckily.) The culprit? LLMs process information by breaking language into tokens – AI building blocks that can be whole words, parts of words, or single letters. They might recognize "straw" and "berry," but counting letters stumps them. With numbers, some treat them like words, comparing 11 to 9 without respecting the decimal point. This reveals a fundamental truth: LLMs are language models at heart, not tiny humans in boxes. They excel at pattern recognition but often stumble over logical reasoning and math. It's like a brilliant literature professor who can dissect Shakespeare but gets confused by a restaurant bill. As we chart the course of AI development, these limitations should inform our approach. We're not building infallible oracles, but powerful tools with distinct strengths and blindspots. The challenge ahead isn't just about making AI smarter, but making it smarter in the right ways – ways that complement rather than attempt to replicate human intelligence.

Nvidia's Eagle: The AI with X-Ray Vision (Almost): In a new research paper, Nvidia has unleashed "Eagle," a new AI system that's less bird-of-prey and more visual savant. This digital Sherlock Holmes doesn't just look at images – it practically dissects them with the precision of a master detective. Eagle can process images in ultra-high definition, spotting details that would make even the most eagle-eyed human jealous (1024x1024 pixels of crystal-clear AI vision). It's like giving a computer a magnifying glass, a photographic memory, and a degree in art history all at once. Eagle also has a whole team of specialized visual "experts” working together – one part is great at spotting objects, another at reading text, and yet another at understanding the overall scene. Together, they create an AI that doesn't just see – it comprehends. The implications are huge. From deciphering doctor's handwriting (a true superpower) to analyzing complex scientific imagery, Eagle could revolutionize how we interact with visual information. It might even make those "I am not a robot" tests challenging for once. Nvidia has decided to release Eagle’s code to the public, making it open source, although they are emphasizing the need for trustworthy AI development, addressing concerns about bias and privacy head-on.

— 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.