What’s your current driving interest in integrating AI into your creative process? For many, it stems from a healthy dose of curiosity mixed with some good old-fashioned fear, uncertainty, and doubt.
That’s only natural given the rapid pace of innovation in AI right now. But after a year of constant change we’ve hit a point where we’re seeing the limits of current consumer-facing AI like chatbots and art generators. Yet we’re still waiting to see if another meteor-sized leap in AI is about to come crashing down even as we’re processing the impact of the one that landed at the beginning of this year.
For those of us already actively using AI, this moment is bringing a sense of relief. Things have slowed enough that there’s time to actually work with the existing tools and explore their creative potential.
Many argue AI quality has actually diminished recently, convinced that by feeding on its own data, the models have become blander and less effective than just a few months ago. But that perception is likely more a byproduct of us becoming accustomed to the technology’s capabilities combined with the common human instinct that in a world of constant change and innovation anything not constantly improving must be actively getting worse.
In truth, for AI to keep advancing as a useful consumer tool, it needs to become more responsive to users’ needs. That’s been the story of software innovation generally, but the current AI technological pivot points are misaligned with the priorities of the customer.
That’s not to say efforts to customize AI aren’t underway. The generative art community working on Stable XD offers an inspiring example, revealing what happens when people have time to more deeply explore a stable tool. It’s not dissimilar to watching game developers squeeze more and more innovation out of a console over its lifetime.
But most focus now remains on improving the core technology itself. That’s understandable given we have yet to fully grasp how AI models gained conversational ability. As we uncover more of those inner workings, we can simplify and target large language models to run efficiently on basic hardware.
While boosting raw processing power offers obvious benefits, an under-emphasized need is adaptability. We require AI systems capable of optimizing dynamically to users’ unique contexts, goals, and feedback. A one-size-fits-all model contains inherent limitations.
One promising approach gaining attention is transfer learning, where models are first trained on massive datasets then fine-tuned to specific domains. By tweaking a generalist foundation, we can create AI tailored to individual use cases.
Applied to creative pursuits, this enables AI to adjust on the fly to artists’ styles and preferences. The machine becomes less a static tool and more an ever-evolving creative companion.
On the research side, there is intense focus now on developing more advanced AI models. Google’s LaMDA made waves as an experimental conversational AI, hinting at interactions approaching human-level — though not without controversy. Meanwhile Stable Diffusion’s open-source image generation model built on robust CLIP embeddings allows creators to locally run GPU-powered AI with impressive artistic capabilities. As these large, specialized models advance, the ability to pack AI into streamlined low-resource or lightweight forms broadens access. The quest continues to enable stable on-device AI ready to assist anyone’s creative goals. But when that’s going to happen still remains an unanswered question.
But customized AI remains a longer-term vision. Present efforts focus more on helping users work within current constraints through techniques like prompt engineering: the art of guiding AI output via carefully formatted text prompts.
Improving those skills can make generating the desired results feel more under the users control. Still, it places the burden on the developer to learn prompting best practices. And because every session of AI is constantly attempting to learn, even those results can seem inconsistent. An ideal scenario would the AI to begin to understand intent and explain that understanding in natural language to get satisfying creative collaborations.
That kind of seamless interaction likely requires advances across training datasets, model architecture, and user interface design. There are also calls to develop AI with a consistent personality to enable more trustworthy back-and-forth dialogue.
But will the world ever be ready to trust a machine that’s explicitly trying to read the mind of its users?
In the meantime, taking time to intentionally experiment with different prompts and settings goes a long way. Maintaining an open, curious mindset allows you to sidestep preconceived limitations. By exploring possibilities, you chart a course rather than getting stuck thinking AI can’t do something.
As I mentioned in my previous post, approaching AI as a partner to creatively problem-solve with rather than just a pixel generator to bark orders at unlocks its potential.
Like it or not, this technology still has far to go. But embracing imagination over expectation can help to inspire more compelling results as AI steadily progresses. And as your hands-on learning feeds real-world data back to researchers, it will inform future improvements.
So while uncertainty around AI persists, don’t let it hold you back. Whether we are sitting in the eye of the storm, or the hurricane of innovation has passed over us leaving us to pick through the aftermath, now may be the perfect time to jump in and see what’s possible.