Tools shape how we do creative work. AI tools need good constraints, high sensitivity and an understanding of remixing to make them easier to wield in the search for great ideas.
* “AI tools are so nascent that they’re not exposing very many knobs to creatives” - part of me wonders whether that’s even possible. I’ve read Wolfram’s paper on ChatGPT, and I could see _him_ be in awe/wonder about the technology behind LLMs. The innards seem so fuzzy and non-deterministic to me.
* “Struggle is an emergent property of doing creative work.” - it rings true and yet something about this statement bothers me. I think it’s because I’ve spent all of my career on helping non-creatives move faster, so anything else screams inefficiency to me. But that’s not true here, I respect that.
* You used music a lot as examples in this piece - I wonder if writing could have been included as well.
2. I'm not exactly sure about exposing the innards of the model. It's all layers of the stack, right? I'm not changing the GPU instruction set, but someone at Apple said, "We should do this now," and was totally right. Where we live, higher up the stack, it might be seeing what the rag results are raw or what prompts are actually going into the context window. On the art side, I have been playing with ideas to explore the embedding space more directly.
3. I see it the same way I see SPC. The struggle isn't building the crud app. It's finding the app worth building. If there wasn't struggle there, a startup would already be working on it. Struggle is finding the edge of knowledge and picking away at it.
4. Writing could totally work. Letters alone seem like a great example. We agree on a set of sounds and build out words from them. There are more sounds than those assigned to letters, but now we're working in a smaller space that's easier to manage. If you want to go up the stack, the hero's journey and save the cat are structures that help tell better stories because they are reasonable constraints.
Totally! I have all these scars from ~2015 when we collectively tried to make chatbots for everything, and it was just so slow. Interfaces are amazing.
Great article. Couple thoughts:
* The guitar and violin analogy works well for me
* “AI tools are so nascent that they’re not exposing very many knobs to creatives” - part of me wonders whether that’s even possible. I’ve read Wolfram’s paper on ChatGPT, and I could see _him_ be in awe/wonder about the technology behind LLMs. The innards seem so fuzzy and non-deterministic to me.
* “Struggle is an emergent property of doing creative work.” - it rings true and yet something about this statement bothers me. I think it’s because I’ve spent all of my career on helping non-creatives move faster, so anything else screams inefficiency to me. But that’s not true here, I respect that.
* You used music a lot as examples in this piece - I wonder if writing could have been included as well.
1. ❤️
2. I'm not exactly sure about exposing the innards of the model. It's all layers of the stack, right? I'm not changing the GPU instruction set, but someone at Apple said, "We should do this now," and was totally right. Where we live, higher up the stack, it might be seeing what the rag results are raw or what prompts are actually going into the context window. On the art side, I have been playing with ideas to explore the embedding space more directly.
3. I see it the same way I see SPC. The struggle isn't building the crud app. It's finding the app worth building. If there wasn't struggle there, a startup would already be working on it. Struggle is finding the edge of knowledge and picking away at it.
4. Writing could totally work. Letters alone seem like a great example. We agree on a set of sounds and build out words from them. There are more sounds than those assigned to letters, but now we're working in a smaller space that's easier to manage. If you want to go up the stack, the hero's journey and save the cat are structures that help tell better stories because they are reasonable constraints.
I like the quote in this article https://www.geoffreylitt.com/2023/03/25/llm-end-user-programming.html
"Chat will never feel like driving a car, no matter how good the bot is". As you identified the input space is too big.
Totally! I have all these scars from ~2015 when we collectively tried to make chatbots for everything, and it was just so slow. Interfaces are amazing.