Building Cambra: a language + database for coding agents. Previously Snowflake, Google, Twitter. Distributed systems, types, and time outside.
Boulder, CO
Joined January 2012
- Tweets343
- Following222
- Followers269
- Likes211
"Verified but Vacuous"[1] empirically studies the difference between:
A) the verifier accepting a proof
B) theorems matching the spec
C) the spec matching real requirements
Unsurprisingly, (A) is much easier to achieve with AI than the others. Getting practical benefits from verification requires all 3.
So how do we do that? IMO, the current conversation misses a crucial element: aligning the abstractions in our code to those of the real world. It's much harder for an AI to sneak an invalidating assumption into a theorem when it must work in terms of application-level concepts.
Reaching this alignment requires software infrastructure to raise the level of abstraction even higher, so every proof doesn't have to handle the myriad low-level guarantees encumbering modern distributed applications.
[1]:arxiv.org/html/2611.05590v1
Dan Sotolongo retweeted
Incremental processing is the foundation of efficient stream processing across the entire latency spectrum, yet SQL lacks a native mechanism for querying changes to a table over time. Today we present our #SIGMOD paper describing Snowflake Change Queries dl.acm.org/doi/10.1145/35897…
Dan Sotolongo retweeted
Question for Data Twitter. What are your most important SQL Window Function use cases?
Rank?
Sequence numbers?
Rolling Averages?
Row Totals?
Cumulative Sums?
What else?
Dan Sotolongo retweeted
Galactic atmospheres influence how galaxies live and die, but how do they work? Join us Friday, December 2 @ 7PM PT for a public astronomy lecture on galactic atmospheres. It will be hosted in-person at Caltech and streamed on Youtube Live. See you there!
youtu.be/xmxW7z7IjfY
The #current2022 talk I gave with @takidau is online! It’s called “Streaming 101 Revisited: A Fresh Hot Take”. We talk about what’s missing in existing models and propose new ideas to make declarative stream processing easier and more powerful than ever.
Dan Sotolongo retweeted
How can we grow beyond the limitations of past #streamprocessing systems? What awaits us when we do? 🤔
If these questions keep you up at night, don’t miss this #Current22 panel featuring @takidau and @sortalongo from @SnowflakeDB.
More info: fal.cn/3rtdK
Very excited about our #VLDB2021 paper on #watermarks. We've observed some confusion around the what, why, and how, so it's an honor to get to discuss them in a research context.
Give it a read if you're interested!
vldb.org/pvldb/vol14/p3135-b…
I left a comment on a PR suggesting adding a TODO linking to a ticket. That’s a lot of levels of delegation to tell myself to do something…
Great conversation about the nature of explanation. I love seeing @DavidDeutschOxf ideas being built upon
Mindscape 150 | Simon DeDeo on How Explanations Work and Why They Sometimes Fail. We run the gamut from serious science to loony conspiracy theories. #MindscapePodcast
preposterousuniverse.com/pod…
watching a Burger King employee shift receipts one by one on the order slider thingy, and all I can think is "circular buffer"
A developing, autocratic country had the spare time and money to organize an international rescue effort of 13 nobodies, and everyone's mad that the billionaire who tried to help insulted a couple of the rescuers. I'm still rubbing my eyes at the first part.
Dan Sotolongo retweeted
URGENT: Huge explosion rocks #Tianjin, northeastern China teapaw.org/thelastnext/status/631… (Video via @Holbornlolz) Details on RT.com
I’m a twitter noob… didn’t .@ teapaw.org/sortalongo/status/6265…
@momv here we find a vivid portrayal of the Malthusian catastrophe: a dramatic rise giving way to a chaotic breakdown
