It’s been a really wild year in the AI industry, and I have some thoughts on this category of technologies and its potential, alongside some heuristics I use when learning and thinking about these sorts of tools.

1. What we’re talking about here is not artificial general intelligence (AGI): it’s unlikely to become conscious, do anything it wasn’t programmed to do, or go Terminator on us. It’s often convincingly lifelike, but so are a lot of things that aren’t “thinking” in the human sense of the term.
2. There are a lot of players in this space, and they have different motivations, backgrounds, and intentions. OpenAI is a big one, backed by wealthy people and tech-world corporations and organizations (it has both nonprofit and profit-seeking divisions), but there’s also Stable Diffusion, which is more open and was developed at a research university in Munich, Chinchilla (which is a spinoff of Google-owned DeepMind), and BLOOM, which is a multi-language model developed through a workshop held by an AI company called HuggingFace.
3. Those things I just listed are all different in terms of conception, capabilities, and the roles they play within this space (some are sub-companies or brands, some are tools, some are systems used by the tools), and understanding that this is a big, varied industry is helpful for understanding these tools, how they apply (and may someday apply) to us and our lives/work, the pros and cons of each, and where these things might go next.
4. Many of these tools are only really accessible to folks who are willing to learn some code and development techniques. Most of the bleeding-edge stuff falls into this pro-tier category, but an increasing number of well-built, everyday-user-accessible work is becoming available, and that’s the stuff most of the press you’re likely to see on this topic focuses on.
5. Those accessible tools are getting very good, very fast. And by “good” I mean interesting, fun, useful for all sorts of things, and simultaneously confounding, worrying, and awe-inspiring.
6. Some of the (currently) most-useful and most-impressive ones (to me, at least) for messing-around purposes are ChatGPT, DALL•E 2, Stable Diffusion (which I usually access via a Mac app called DiffusionBee), and Midjourney—though there are a lot more out there and new ones are popping up every day. Most of these are free to us at the moment, though they typically require you sign up for a free account and you’re usually limited in terms of how many “things” you can make before being cut-off in a given period of time.
7. Those “things” include images (made from text prompts you provide or by riffing on existing images), writings (essays, poems, explainers, songs, screenplays, games, etc), code (websites, apps, etc), and increasingly other media like animations, music, and videos.
8. The quality of the things these AI tools make for you vary substantially based on the tool you use, the quality of your prompts, and other such variables. I’ve personally found that after messing around with these tools for a few minutes each day, my output (what they make for me) has gotten a lot better (in the sense of being more interesting and more useful for my intended purposes).
9. Much of the information these things provide will be incorrect, and fact-checking and editing is often required. They’re good at projecting confidence, but not so good at backing that confidence up with consistently accurate info. You’ve been warned.
10. Also, there are significant ethical questions inherent in these tools, as they basically pull from existing writing and imagery to cobble-together new-seeming things (this is an ultra-simplified explanation of what they do). That means they “borrow” liberally from huge libraries of existing work, which at times means brazenly ripping off artists, writers, and other makers who aren’t compensated for their contributions to these new works. We’re still collectively working through whether and to what degree this is okay and what can be done about it if not, but keep that in mind as you experiment.
11. In terms of utility, I mostly use these tools for inspiration, and I’ve thus far been able to use them to develop a few simple web apps, to generate an array of visual inspiration for a logo I designed, and to churn out some outlines for potential long-form writing (which helped me come up with alternative ways of arranging the information I wanted to present).
12. The term “endless media,” in this context, refers to the possibility that these tools (many of which are predicated on what’re called “large language models,” like GPT-3), could overwhelm all media channels, churning out an infinite volume of images, writing, videos, and games because such content is suddenly so easy and cheap to make at large scales. Consequently, we could see a whole lot more of (AI-produced) everything, and anyone who makes such work in the conventional way might be nudged out, no longer able to produce on the necessary scale and at the necessary cadence.
13. This is a real concern for both makers and consumers of such work, and like any potentially revolutionary new tool (Photoshop when it was first released comes to mind), it’s likely that some of these predictions will come true, others will prove to be overblown, and the most revolutionary changes will be those we don’t see coming.
14. My heuristic for using and thinking about these sorts of tools—sharp ones that can possibly empower me in exciting new ways, while also being dangerous if wielded without intention and forethought—is to learn what I can about them, do my best to stay excited about their potential (rather than exclusively frightened of their implications), and to maintain a situational awareness that helps me integrate what works, dismiss what doesn’t, and hone my capacity to use them over time.
15. At the moment, for me, these tools are primarily means of generating inspiration, helping me think along different paths and from different perspectives than usual, and to provide visual or informational inspiration before I dig in and do things more or less in the usual way. I’m open to that changing, and I would truly love to see these things evolve into maker-focused augmentations, empowering all of us to do more and better in less time (and with less effort)—but I’m also wary of changes that would lead to an abundance of empty, derivative fluff that fills-out content libraries on the cheap, but which robs real-deal makers of their livelihoods in the process, and which doesn’t otherwise contribute to individual or social growth.

(This episode of Let’s Know Things focuses on this topic, as well, though from a more informational and less editorial perspective.)

If you found value in this essay, consider buying me a coffee :)