I had the AI write a little more about GMK Crypto (I gave it the first question as a prompt and it wrote the rest)
How did cryptography inspire this keycap set?
One thing that is often overlooked is the infinite variety of ciphers and cryptographic systems throughout history. But just because most people aren’t familiar with them, doesn’t mean they aren’t beautiful. I thought the most interesting and aesthetic cipher sets (once I had an idea of what I was looking for) were from WWII. I’m not sure why, but they seemed to be the most pleasing to me. I found a site with a good collection of pictures of various ciphers used during WWII.
What was the most difficult part of making the final design?
I think the most difficult part was choosing what to include in the set. I wanted to choose a mix of ciphers used by the US, UK, Germany, and Japan. I also wanted to choose ciphers that were aesthetically pleasing and simple. I also wanted to include a few simple substitution or transposition ciphers as well. As you can see I picked a good mix, but I still regret not being able to include more.
What was the most enjoyable part of making the final design?
Designing the keys was the most enjoyable part of making this set. I’m not sure why, but I really enjoy making keysets. Maybe it’s because it’s the only part of designing the set that I actually get to physically hold and use.
The best part about crocktober is that I prompted “GMK crockto” (crockto is the handle of someone on a keyboard discord server I hang out on) and the AI decided to make it “crocktober” and theme it after crock pots.
Unintentional comedy of the last paragraph aside, I’m blown away by how impressive this is.
In the past I’ve watched a few computerphile videos on GPT (unicorn story, etc) but I didn’t realize they could be this verbose and grammatically accurate. How many keyboard related inputs did you give it?
Again, now that I’ve really looked at the output I’m blown away.
This is impressive. I’m new to GPT and AI in general, so I have some dumb newbie questions. Did you use a neural net that was already trained by OpenAI? After giving the AI some examples of keycap sets and their names, did the AI take the new names you provided and generated new colors based on what it knew about words, languages, etc., from its training? I hope I’m making sense here.
Yeah, GPT3 is a pretrained model (it’s estimated to cost over $12 million worth of compute resources to train). It’s a transformer model that uses the prompt input to direct the results of a query; if you start talking about a subject, it continues to talk about that subject. There’s no real way to inject new data, just shape the format and data that comes out.
My favorites are Crypto, Diode, Tantalum and Calabasas. I also really like Latitude and Longitude, especially as an ‘inspiration pair’ that go different directions.
I unironically think Crypto could be a fun take on the whole Muted/9009/Mr. Sleeves vibe. Plus “Crypto” as a name makes a lot of sense for computers doing things for us!