I worked with my kids for a couple days and built a card game using ChatGPT, Midjourney and some other AI tools. One of a couple AI side projects I did over the holiday break.
Here’s how:
agree. Am just finishing this book recommended by @bgurley that’s been a treat. Storytelling critical to raise $ but also to recruit and effectively lead any org. https://amazon.com/Storytelling-Animal-Stories-Make-Human/dp/0544002342/ref=nodl_?dplnkId=c7d0a9b0-ab3c-44c6-80a0-0de490abbb96…
Recorded a bit of the first public @Cruise ride from my perspective.
Even in SF at midnight we saw two illegal u-turns, a construction detour, and an ambulance, which our car (“Buggie”) navigated perfectly.
Games have been solving the synchronous social problem by quietly adding AI bots. Big betray of trust for players when found out.
There’s a new realtime social network waiting to be born that’s more PvE. The AI is fun to talk to alone, but better together.
Live apps like Clubhouse are hard to bootstrap because you need users simultaneously. Now you can AI generate "people" to keep users interested until enough humans are online. Diabolical, but would likely work.
This is true.. actually the best first gen ideas will take advantage of hallucinations more than trying to fight them.
On the path to making LLMs 100% right, let's not forget the power of weird.
(also.. doing self driving was the right call, but new things require patience)
I’m seeing many fall into the “self-driving trap” w/Gen AI
The self-driving trap is seeing shiny demos & thinking said demos will reach 100% reliability for prod w/in a few years
Many proposed Gen AI use cases need 100% reliability, and thinking that’ll come soon is a mistake
Many startups are still designing user-facing AI in a pre-ChatGPT style. If you added a single text line prompt w/expectations of bullseye results it’ll disappoint. Users weant to wander the idea maze.
Interfaces should encourage co-op w/AI. Aka
“yes that, with more cowbell”
It’s been fun to dig in with the kids on a project we’ve talked about for a long time.
And as always, really using these products to produce something finished vs just tinkering exposes a lot of nuance that hones your thinking on where this should head next.
Many of the ideas were pretty average, but that’s not surprising for an LLM that is predicting the *average* next phrase/image/etc.
AI got ideas on the page instantly we could improve on. There is a lot of playtesting (these are not final mechanics!) but we’ve got something.
And after a couple days we had a working prototype full game!
AI felt most valuable at solving the blank page problem.
“Ideas for themes?”
“Show me 10 cards with that idea”
“Titles of mining personnel on a spaceship”
“How would you group these 60 cards into classes”
We also experimented with Scenario to create card icons, Anthropic/Claude for characters & card class groupings (which outperformed OpenAI in this task), some Hugging Face models, and a few other AI tools along the way that haven’t quite made it into the game yet.
We used the box design image to prime Stable Diffusion, MJ, and a Hugging Face model to do a bake-off for a logo. Then used a font matcher, bought the font we liked to match the style, and a little photoshop to clean it up.
For the box design, I've been watching MJ for a long time and had fallen in love with this kind of pop art, high contrast look so I used that to create some good images with movement.
We used one anchor image w/MJ to generate the other card art in the same style. This took the most time, as current models are iffy at context. Often it forced the same character, not just style. And it loved guns & the same hair style despite heavy negative prompts.
We picked sci-fi. In the future warring cities (SF, NY, Beijing) each launch a space-ship trying to reach the “nearby” Kepler 425b first.
From here we iterated on a style for the art in Midjourney & Lexica, with the constraint of not using “in style of” any particular artist.
My kids and I love drafting card games (Dominion, Star Realms, Ascension) but each have flaws we wanted to fix.
ChatGPT got us quickly brainstorming on themes from medieval fantasy to 20s mafia. Then we used Claude + Fermat to test iterations of ideas on cards & mechanics.
also just updated Spaces to have Discord-bot support.
If you’re building an AI startup and not thinking about launching via Discord, here’s why you probably should.
👇🧵
Excellent from Scott.
Compensation/respect will increasingly come from the uniqueness of the idea vs the technical skill of its execution, a 50+ yr trend.
If the net confidence of humanity to feel creative grows, amazing.
Hard not to feel worried about the disruption though.
Took some time to share early thoughts on (1) the death of creativity’s learning curve, (2) the opportunity for creative pros in the era of creative confidence, and (3) implications for creative careers, culture, & beyond...
https://scottbelsky.medium.com/creating-in-the-era-of-creative-confidence-b4e251d725f…
Some founders are built for puzzles.
Some founders are built for mysteries.
Pick wisely. Many succeed in one and stumble badly into the other. I’m not sure if it is because they are not willing or not able to play a different game.
It only took 58 years to go from the first chatbot (ELIZA, 1964) to one that really works (ChatGPT, 2022). Don't confuse a clear vision with a short distance. And don't confuse a long distance with impossibility.
Just had someone ask why I joined the AI avatar trend. Same thing happened when folks thought I had an NFT avatar.
Truth is it’s neither, just something awesome
this morning…
- What % of college essays next cycle will be written by AI?
- Shouldn’t it be all of them?
College essays are not a creative exercise, they are a production function for writer & reader. Three formats & a few variables changed per student
Who wants to take an under-over on the percentage of college essays submitted this cycle and next cycle that will be written by AI twitter.com/Afinetheorem/s…
This is not plant based salmon, it’s not wild or farmed. It’s cultivated salmon. Made from growing salmon cell by cell. And it really blew my mind. So tasty and much better than previous versions I’ve tried elsewhere.
Thank you
Realized I love #Andor because Star Wars finally skipped the tired “space opera” and made a classic Sci-Fi banger.
A what if scenario, in unfamiliar setting, change a few key technologies w/societal impacts, then relatable characters react to that society. Classic Sci-fi 🪐
Today @Cruise launched *daytime* driverless rides. Service is live for Cruisers and will launch to the public soon.
First ride was this morning around 8:30 am:
. Testing the new release for video editing has been magical.
Such a seemingly simple set of UX advances that make integrating AI into my audio/video workflow dead simple. 👏
Today is a really big day at Descript
First, the all-new version of Descript is now available to everyone. Second, we're announcing that the @OpenAI Startup Fund is leading Descript’s $50 million Series C.
Details here: https://descri.pt/3AfoVrl
Or keep reading
1/21
someone asked me why i use AI when i'm a skilled painter. isn't the joy of art in the process? not for me. as a disabled artist, it's physically painful. there's no joy in it. the concept is what i love, envisioning & bringing it to life. AI enables me to do this without pain.
Just have to say, given this week. There are a lot of wonderful founders who do not obsess about being the *main character* with a godlike complex. Many of them.
They have perhaps more ambition, but more properly focused. Kudos to them, they win in the long run. ❤️ you all.
Tonight our driverless service area in SF expands to cover almost all of SF. Today is also exactly one year since my first ride, the first driverless ride in a major US city.