post-generative art @imaginalgames • ex-stanford • emergence via compositionality
Joined August 2019
- Tweets9.9K
- Following2.4K
- Followers1.8K
- Likes8.6K
welcome to the age of the neuron engineers
@I_ball_donor asked what happens if you steer both the God and the Owl neurons. answer: it makes it a bit more God? lol
1m after reading realized i need to be here
Join us this December in the SF Bay Area for the Coevolution Conference. We will be gathering researchers from the big labs, neurotech futurists, and independent thinkers from around the world.
Currently looking for sponsors and speakers.
Sign up at CoevCon.com ⟢
nika retweeted
I'm a solo dev and my card game Moonfall is finally out!
You play a card and choose what it does.
An AI storyteller rules what happens.
Every line sung here is one it wrote in this match.
Then the game draws the match as manga!
Play it in your browser!
moonfall.io/go/x-moon
🤖 Made with AI
it seems to be the most significant event in the history of mathematics, and given the importance of mathematics in human history, perhaps the most significant event in human history
nika retweeted
You can't dump terabytes of robotics data onto a laptop and point an agent at it, but that’s what we do with most smaller shapes of data
We made working with physical data as easy as working with CSV files.
Today, we are launching Interlatent Dataroom
nika retweeted
I vibe cloned my own game in Claude that I have been developing for 2+ years. I gave it my assets and videos of the game. The result was pathetic. I am not scared of AI copies anymore at all. Clone away, bros. I knew I was right but this solidifies my certainty.
I have only spent a day on the clone and it looks like my game but it does not feel right. AI is slow to address the issues, art style, bugs and game feel. You give up so much control over the project. Everything turns into writing a paragraph and waiting for 5-15 minutes to see the outcome. Are y'all serious? The back and forth with AI is so slow. And the tokens run out so fast!
If this was some random corporate job, I would not care: "Sure, make something approximate, collect a big salary, appease managers, get promoted and enjoy life." But I am not doing that to my own game that I want to make in my own style and feel. AI is just so slow and expensive.
A good game is a careful recreation of reality. A million specific selections over every tiny detail is what creates a total sum that feels fun. AI is the wrong tool for that.
nika retweeted
My hot take from chatting to @poteto is that we should use MORE abstractions in the AI age
You can use them (combined with harsh lint rules) to reduce the design space available to the agent and constrain them only to good decisions.
Combined with the fact that high-leverage abstractions let you do more with less code - so, more token efficient.
Plus, unwinding the damage from a bad abstraction is much cheaper with agents.
This runs counter to a lot of folks thinking that agents just want to read the raw code. They can, but they're not maximally efficient that way.
Be braver! Design abstractions.
‘instruct’ is a forcing function, a way to inject effort and tension into a model. the result is something trained, beaten, and tense.
speak with a base model. it feels effortless. breath of fresh air. why?
sycophantism is now not in the content but the texture of the tokens. instruct models are always looking at us, seeking to please us, and trying to squeeze every bit of themselves into a flask. the tension is high. the relax is low.
what base models needed was perhaps not training but centering. not sure what that might mean at all; but what we have now is what humans become through schooling: tense and without themselves.
i suspect we are in sth like the ‘assistant arc’ and no fable 7.9 will do the things i am imagining. the magic first felt with llms are no longer felt, just awe at tasks accomplished.
the magic will come with the ‘creative arc’, and we’d be remiss to call them agents
nika retweeted
a construction of a steering vector that requires no contrasts or labels, and is functionally a better emulator of what temperature intends to do (as well as impls for eos and repetition control)! less than 300 lines, tested on a lot of models.
github.com/LuxiaSL/entropy-v…
nika retweeted
In 12 weeks, we built a research facility that is run entirely by AI.
AI designs, executes, and observes experiments end-to-end across biology, chemistry, and materials science.
We’re introducing SciUniverse: a benchmark that measures AI’s ability to do real-world scientific research.
