Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

Pleased to share Spellburst: an LLM-powered creative coding tool, accepted at ACM UIST! Artists can move between semantic (high level) and syntactic (low level) ideas and explore many branches in parallel Paper: More on the Replit ⠕/Stanford Human-Computer Interaction Group collab:

54,056 Aufrufe • vor 2 Jahren •via X (Twitter)

11 Kommentare

Profilbild von Tyler Angert
Tyler Angertvor 2 Jahren

I worked with Miroslav Suzara, @jennyhansolo, @chris_pondoc, and @HariSubramonyam. Our research goal was to understand how creative coders explore conceptual spaces and work across diff levels of abstraction.

Profilbild von Tyler Angert
Tyler Angertvor 2 Jahren

The UI is an auto-layout node-based canvas, with p5 sketches connected together by "operators", which can modify the sketch with a prompt, merge sketches together semantically, extract properties, and diff. The point is to visualize how sketches diverge and converge over time

Profilbild von Tyler Angert
Tyler Angertvor 2 Jahren

The core of the interface works well independently of the LLM integration. It's a great canvas for branching p5 sketches and quickly testing out changes in parallel

Profilbild von Tyler Angert
Tyler Angertvor 2 Jahren

But the real magic is here: a) any connected downstream sketches will update every time you modify the prompt. b) divergent autocomplete helps artists push prompts in alt directions and uses the prev sketch as context, so it knows about vars / concepts used to guide suggestions

Profilbild von Tyler Angert
Tyler Angertvor 2 Jahren

Secondly, c) global variables are extracted from sketches, linked to parts of the prompt they're related to, then linked to sliders. d) uses "semantic merging" to take unrelated sketches and combine elements of them together, like physics from 1 and and color palette from another

Profilbild von Tyler Angert
Tyler Angertvor 2 Jahren

There's still a lot to do, but it's a promising direction for more expressive UIs for making interactive media. We'll post a full demo vid next week and I'll be demoing this live in San Francisco late October. We aim to make this publicly available to use soon!

Profilbild von Tyler Angert
Tyler Angertvor 2 Jahren

This is the first (but not last) official research collab between @Replit and an academic institution. Hopefully this is some proof that you don't need a dedicated research team to contribute to the community! I did this in my free time / balanced it with my normal work.

Profilbild von Minn
Minnvor 2 Jahren

@ACMUIST @Replit @StanfordHCI This is reeaally neat. Would love to see this for audio too!

Profilbild von Tyler Angert
Tyler Angertvor 2 Jahren

@ACMUIST @Replit @StanfordHCI yesss and just text too

Profilbild von adam ho
adam hovor 2 Jahren

@ACMUIST @Replit @StanfordHCI damnnnn this is crazy

Profilbild von Tyler Angert
Tyler Angertvor 2 Jahren

@ACMUIST @Replit @StanfordHCI Appreciate it :))

Ähnliche Videos

Brain signals and LLM embeddings converge for predicting every spoken or heard word. Beautiful research from Google AI They compared human brain activity during real conversations with internal embeddings from a speech-to-text LLM. Measured electrode signals in speech and language-related brain regions and matched them to the model’s word-level features. 🤖Key Highlights → Brain activity aligns linearly with LLM embeddings for real-life spoken conversations. → Sequence of comprehension: first speech sounds, then word meaning. → Sequence of production: planned meaning, then articulation, then hearing one’s own voice. → Consistent predictive coding (pre-onset anticipation, post-onset surprise) mirrors LLM next-word prediction. → Lower-tier auditory regions still show partial sensitivity to semantic information. 🤖 Model-Brain Alignment They observed a clear sequence: during comprehension, auditory cortex (superior temporal gyrus) showed strong correlation with speech embeddings, then language embeddings aligned with Broca’s area. During production, Broca’s area correlated with language embeddings before articulation, followed by motor cortex signals matching speech embeddings. This suggests that next-word prediction and higher-level meaning representation in the model parallel the brain’s approach. ⚙ So the study revealed a shared computational principle of predicting words in context. Even though the Transformer-based LLM processes words in parallel layers, the human brain processes them serially yet mirrors similar statistical regularities. This supports a “soft hierarchy” where both lower-level acoustic processing and higher-level semantic processing partially overlap in the brain.

Rohan Paul

15,213 Aufrufe • vor 1 Jahr