Today we’re launching Fieldwork, FLORA’s new home for creative AI education. Every creative we talk to has already decided to learn AI. What they’re still asking is where to start, and who to trust. The loudest corner of AI runs on stunts, like launch videos cut for the algorithm and promises of unlimited everything. None of that teaches anyone how to do the work. Creatives have always learned from other creatives, so every Fieldwork instructor is a working one. They spend their weeks inside agency and in-house teams figuring out where AI fits in real production.

design work on our team is completely changing we don’t spend time producing the work anymore, ai does that. We just make the decisions we gave ai our brand guidelines and let it explore all night, and i really like what it came up with here is how this happened ↓

This talk is a beautiful articulation of one of the most important design problems of our time. A few thoughts on planning with agents: - I'm finding one of the best inputs to plans are recordings of live conversations among a team. this tends to capture a *ton* of peripheral useful context, that's hard to capture even in a solo voice recording -- because people elicit intent from each other. A generalization of this principle is that agents thrive on broad, high level, "messy" context - just like senior employees. - I think visual outputs from agents are critical to complement text, but I'm actually of mixed minds on interaction. Yes, it's clearly nice for certain kinds of things like twiddling visual parameters. But most of the interactivity my agents produce feels mostly designed for dopamine reasons and not providing much utility. I also think interaction can inhibit deep thought compared to linear reading. I actually think for many deeper problems I would prefer top notch prose -- Richard Dawkins / Steven Pinker / Annie Dillard level -- from my agents over any kind of flashier output. - One mental model I find consistently helpful for planning: the job is to establish a *design space* together between human and agent -- that is, a set of relevant dimensions of variation -- and then to efficiently explore the space. Getting the dimensions of the space right is actually one of the hardest parts. Once you have that, you can ask an agent to plant a bunch of starting points spread throughout the space, and then locally hillclimb from the best one.

2 years ago, I started exploring "free will in media". Could you make games that let players explore beyond the designer's imagination?

fun hack, GPT-2 inference running on my jailbroken Kindle Scribe this is real speed, like ~13 tok/s at 4-bit quant of the 124M sized model

Few people could help us understand such current design problems with as much respect for historical & human context as @Mappletons. In addition to being generally great, I also appreciated that I ended with more questions than I began, which accurately reflects how early we are.

Drawing clouds in the sky

Everything is too close to everything

Good exploration of timelines and physics-like hierarchy in UI inkandswitch.com/ink/notes/c…

I recently gave a kinda personal talk at @recursecenter, where I shared some stories about taking your interests very seriously. Watch on YT → piped.video/MccJdr61xnc?is=Pzwm…

but its not just aggregate data vis, its also crucial to zoom into a single trajectory and quickly get a handle on it. inspired by @MatthewWSiu i'm exploring a vertical timeline and expanding messages based on "lenses". perf is key tho, instant scrolling 5k msgs!

i want new ways to look at lots of trajectories, currently exploring: - treemap laid out by groupings (metadata or computed) - colored by classification of success (or grouping) - sized by length - zoom in smoothly to see turns as beads on a string - extremely fast transcript viewer (instant for many thousands of msgs) - switch to 3d to see the forrest for the trees - transition to semantic (UMAP) layout - color by semantic clustering

It would be a tragedy for embedding models to become narrowly optimized for retrieval in the same way that generative text models have become narrowly optimized for conversational rollout. There is much magic to explore in the realm of representations.

More DAGs should be drawn this way. In a pipeline like the one below, the nodes form a list, so each node can be a table row with metadata beside it. Parallel branches are the main limit, but horizontal space is cheap!

I stood directly next to @minafahmi in a room of ~100 people last night as he whispered into his Stream. I couldn't hear a thing he said but watched perfect, immediate dictation appear on his phone — and then moments later on his macbook without him reconnecting anything. magic