2 min read

San Francisco, 2018: What the City Taught Me About AI That No Conference Could

The ambient conversations alone changed how I thought. You can't get that remotely.

I've been here for a few months now and I want to try to articulate something that's hard to put into words without sounding like a brochure.

The thing about San Francisco in 2018 is the density of people working seriously on hard technical problems, and how that changes what feels normal. In most places, saying you're building a machine learning product gets you a polite nod and a question about what machine learning is. Here it gets you into a specific conversation about which approach you're taking and why.

That shift in baseline is more useful than any conference I've attended.

What I'm actually building

Zillion Pitches. The idea is AI-assisted pitch analysis for founders. You record your pitch, the system analyzes it: speech patterns, pacing, sentiment, structure. The goal is to give founders feedback that would normally require an experienced human listener.

The technical core is natural language processing and speech analysis. We're using IBM Watson as the primary ML layer for voice-to-text and sentiment, and building custom logic on top for the pitch-specific stuff.

It's early. The product works well enough to show people and get useful reactions.

What being here changes

I keep having conversations I couldn't have had from Turkey. Not because the people aren't smart elsewhere, they are, but because the concentration of people who have already tried versions of what you're building, already hit the walls you're about to hit, and are willing to tell you about it honestly is different here.

The Draper network helps. Being around founders at different stages of building means you're constantly calibrating against what's real versus what sounds good.

What I'm watching

OpenAI published GPT this year. The language model results are striking. The ability to generate coherent text at length is something I've been thinking about in the context of what it means for our pitch analysis use case. If language models keep improving at this rate, the NLP component of what we're building will look very different in two years.

Worth paying attention to.

With gusto, Fatih.