Bits

Three filters, in order: great people, hard problems, new things to learn. Exceptional people have already cleared the easy problems, and proximity to them selects for harder work. Hard problems can't be solved with what you already know, so learning stops being optional. Compounding knowledge is how you get back into rooms with better people.

Steve Jobs:

Marketing is about values. This is a very complicated world. This is a very noisy world. And we're not going to get a chance to get people to remember much about us. No company is. And so we have to be really clear what we want them to know about us.

Jobs said this in 1997 to a room of skeptical developers. The world he described as "very noisy" would look monastic by today's standards.

There are two kinds of readers: those trying to understand the world, and those trying to understand themselves.

Every now and then, I still catch myself typing in reader.google.com. Old habits.

When I'm struggling to understand something, I break it down. Not once — recursively. Large abstractions first, then the layer beneath, then the one beneath that, until I hit the primal forms. You don't get to understand the thing until you can put it back together.

Software commoditization was the first wave. Hardware is next.

3D printing, CNC automation, and robotic assembly are compressing the cost of physical production the same way cloud and open source compressed software a decade ago. Close that loop with generative models and you get systems that can design, simulate, and iterate on physical goods at something approaching software velocity. The cost floor drops. The barrier to replication drops with it.

Physics doesn't care that atoms are involved.

Secret enemy hideout.

March was a blur — AppWorld, GTC, RSA, and Arm. Four keynotes, a panel, dozens of customer meetings, and tons of high-fives and fist-bumps. The highlight was getting to interview Steve Wozniak. Looking forward to a few days of recharging with hikes and photography in the PNW.

The path I'm on was paved by people who didn't have to help me, but did anyway. That's not lost on me.

I lose myself in the work to find myself in the creation.

Some weeks I'm training a transformer on my iMessage corpus. Others I'm deep in a PNW forest with a camera, or chasing a chord progression late at night. I come to all of it as a beginner. That's not a hobby posture, it's the only way I know how to stay honest with myself.

Far Caspian — "Autofiction" (Live at Leeds Museum):

And this sinks so calcified
Those calluses begin
The last remaining light
Is stretched across your skin

Love this live performance of the recent record. "The Sound of Changing Place" is probably my favorite — the vocal harmonies are beautiful.

Recorded a version of Cold Like Winter on the the OP-XY tonight. I wrote the song more than a decade ago. Something fun about bringing it into the current moment with modern gear. Almost like covering myself.

The Marias — "No One Noticed" (Extended Spanish):

Estás tan dentro de mí
Te sigo pensando
Te sigo esperando
Y estás, oh
Tan lejos de mí, oh
Te sigo pensando
Me canso llorando

The lyrics are from the Spanish outro in this version of the song.

I oscillate between who I am, and who I want to be.

King Krule — "Out Getting Ribs" (Church of Nobody):

And hate runs through my blood
Well my tongue was in love
But my heart was left above
I've got to be leaving now
I thought I'd never be shot down
But girl I'm black and blue
So beaten down for you
Well I'm beaten down in bloom

I genuinely wonder how someone makes a song like this.

Fontaines D.C. — "You Said" (Live from A Night at Montrose, Dublin):

I got into this band last year, especially during M&A diligence — they were on constant rotation. For this show, the band played the entire record front to back and the cinematography is incredible. This was a standout performance for me.

The Voidz — "The Outro" – Secret Show, Live in Venice:

On the back of all the things you gave
If only I was zooming on the moon
I've been in Los Angeles too long I think
Singing songs I never thought I would
Singing in a style I didn't think I'd ever understand
Take it like a California breeze
Grew up on the car horns and city noise
Didn't know the quiet oh oh

I remember walking around my neighborhood and listening to this setlist for months during the pandemic. There's something about these lyrics that hit differently knowing what came next. The LA references, the sense of being somewhere too long, singing in ways you didn't expect — it all feels like the end of an era captured in real time.

Four Tet — Live at Sydney Opera House:

I came to Four Tet pretty late, and this set was my introduction. It's one of those performances I keep coming back to — just put it on and let it run. There's something about the setting, the sound, and the way the whole thing builds that makes it feel less like a DJ set and more like a single long piece of music. A perfect starting point if you've never listened.

Dieter Rams:

Limit everything to the essential but do not remove the poetry.

This is the hardest balance in any craft. Stripping things down is easy — you just cut. But knowing what to protect, the parts that give something its soul, that takes real taste. The best products, the best writing, the best code — they all feel simple, but there's a warmth to them. Something human made it through the editing process.

Mode collapse:

It occurs when the model produces outputs that are less diverse than expected, effectively "collapsing" to generate only a few modes of the data distribution while ignoring others. This phenomenon undermines the goal of generative models to capture the full diversity of the training data.

A few things that stand out to me about this:

• It's not just a model problem — people mode collapse too. You settle into the same patterns, the same solutions, the same responses to situations, and stop exploring the full range of what's possible.
• The most interesting work comes from the edges of the distribution, not the center. When a model (or a person) only produces "safe" outputs, you lose the unexpected ideas that actually move things forward.
• Diversity of input matters. If you only consume the same sources, talk to the same people, and work on the same kinds of problems, you're training yourself toward collapse.