4 ms·

On an emotional level, I totally agree. But as a developer of creative tools, I've been trying to make sense of what's good and bad in this field. Arguably, all

On an emotional level, I totally agree. But as a developer of creative tools, I've been trying to make sense of what's good and bad in this field. Arguably, all creative software is at least partially automating creativity (photoshop, ableton etc). But my conclusion so far is that the gross part is when the user inputs a vague request and a finished product comes out. Contrastingly, if the user can make continuous changes with realtime monitoring of their actions, then that is an instrument - mostly amplifying the creative impulses of its user (which is good).

dalbasal 4y ago
It usually trends to the way over time. When you're demonstrating or studying a tech (like we're doing with gpt now), there's a tendency to go "here's what it can do." Once people start using it for more mundane work, the focus becomes "here's what I can do with it." It probably doesn't matter what children's stories language models can write themselves. It matters what stories people can write using them. Same for software/graphic design/law/etc. These are just tech demos. This becomes clearer once demos play out. 2 years ago, scripting a philosophical debate using language models felt like AIs expressing their opinion. By now, we're past that.
- svantana 4y ago
  I think (and hope) you're right. Recently, someone coined the phrase "bach faucet" to describe the mundanity of an "AI" that can compose millions of novel bach fugues that are indistinguishable from the real thing [1]. At the end of the day that is just pointless - abundant art serves no purpose in society, and its value will be zero when considering supply and demand. [1] https://twitter.com/galaxykate/status/1583907942834716672 https://twitter.com/galaxykate/status/1583907942834716672
  - dalbasal 4y ago
    Precisely. OTOH... using AI to create art shifts around supply and demand such that there will be practical implications. The way email correspondence between A customer and a bank may look very different once both are using GPT enabled email clients.

dalbasal 4y ago

It usually trends to the way over time. When you're demonstrating or studying a tech (like we're doing with gpt now), there's a tendency to go "here's what it can do." Once people start using it for more mundane work, the focus becomes "here's what I can do with it." It probably doesn't matter what children's stories language models can write themselves. It matters what stories people can write using them. Same for software/graphic design/law/etc. These are just tech demos. This becomes clearer once demos play out. 2 years ago, scripting a philosophical debate using language models felt like AIs expressing their opinion. By now, we're past that.

svantana 4y ago
I think (and hope) you're right. Recently, someone coined the phrase "bach faucet" to describe the mundanity of an "AI" that can compose millions of novel bach fugues that are indistinguishable from the real thing [1]. At the end of the day that is just pointless - abundant art serves no purpose in society, and its value will be zero when considering supply and demand. [1] https://twitter.com/galaxykate/status/1583907942834716672 https://twitter.com/galaxykate/status/1583907942834716672
- dalbasal 4y ago
  Precisely. OTOH... using AI to create art shifts around supply and demand such that there will be practical implications. The way email correspondence between A customer and a bank may look very different once both are using GPT enabled email clients.