I've been invited to give an industry keynote at AIDB 2026, and will be attending VLDB 2026 in Boston next week. Thanks to my fortunate connection to Tim Kraska, and Redshift, I do expect to see a few familiar faces there. If you are around, you likely don't know me, but if you do see me, I'd love to talk! I've always been just outside any specific research community, wandering like a vagabond from signal processing, to information theory, to sensor networks, data mining, personalization, machine learning, databases, AI and systems. I've never really had a home conference, maybe this is my chance before they figure out what a mistake they've made. No take-backs Kostas, Subru and Thaleia.
I initially wasn't really sure what I'd talk about. I haven't given a talk at a conference in more than 10 years - though I've had quite a few papers. I always found some excuse to ask a co-author to present, or refuse to travel. Bad idea as it turns out. Everything atrophies in the comfort zone.
I did what any silicon blooded AI pilled pseudo-techie would do - I turned to my only true companion - chatgpt of course. I gave it a bunch of my papers and high falootin' ideas and prayed to the GPU gods. Sol plugged away at it for hours consuming millions of tokens, hours of my time, gigawatts of power and generated 100s of pages for me, of which I couldn't even sit through a paragraph. Exactly like Arvind Narayanan says, I turned to AI when I didn't have much to say and got the hollow writing he predicted. Arvind is rapidly becoming my prophet as I navigate this new AI world, and I do hope to meet him someday and see if he is indeed just chatgpt 6.
Harkening back to my PhD days, I tried to do a staple-gun talk - putting together a bunch of slides from some recruiting talks I recently gave at CMU and MIT. This went a bit better, but not really by all that much. Spoiler alert - starting your writing with AI is a bad idea. Fortunately, despite my disposition, I have friends! Well at least one. Vyas read the 40 odd slides with no coherent story and in minutes gave me the title, the theme and how to fix it. Maybe there is a place for humans after all. While it isn't perfect, or maybe even any good at least it's mine/his.
So, what is the talk about? It's about 'embracing the gap'. On the one hand, AI is taking all the jobs. On the other, as you can probably tell, even I don't know how to do my job, or even what it really is. As Tyler Cowen says, some jobs are messy. My job, at least as I see it, is to 'embrace the gap' and make messy things appear less so to our customers. Maybe I'm safe. I'm going to talk about how trying to build reliable, useful, usable, efficient and simple looking systems out of messy building blocks is hard, interesting, research-worthy and unsolved. Maybe I should re-read John Gall. I'll post more about the talk in a bit once Vyas reads the next draft. Or, I'll post a completely different idea - 'cause messy is how I roll.