one of my favorite ways to read slowly as of late has been to read only until I find something to think about it takes a lot less time to get something out of what you read. higher utility density per word, and you can make sure you don't miss anything

the real problem with lack of sleep for me is not that my intelligence or energy are suboptimal. it's that my ability to stay mindful plummets. I'm generally mindful for less of the day, and even when I am, I'm not able to muster as much awareness as I am when well-rested.

Hello everyone! Excited to share what we've been working on for the last few months: an open source framework for prompting LLMs for analytics. Repo: github.com/oxy-hq/oxy Read more here: www.oxy.tech/blog/introdu...

so much sass from 4o

Without introspection, everything is a waste of time. With introspection, nothing is a waste of time.

People will often tell you that building a company is about grit, about execution, about long hours and sacrifice. But I've been reflecting on this lately, and I think it's generally incorrect.

does anyone know of any speed benchmark comparisons between duckdb and standard warehouses (e.g. bigquery / snowflake) execution times? remember some floating around last year, but can't find them for the life of me

I nearly switched to iphone this year, but the one thing keeping me: Gemini + Kindle is an exceptional user experience. It has fundamentally changed how I read, particularly for esoterica Tap hold bottom bar -> select confusing section -> get answer from Gemini

Me: the worst analysts are zero value-add Friend: no, the worst analysts are negative value-add Anyone have any horror stories? Trying to figure out how bad a bad data team can actually be

imagine if the real modern data stack was oracle all along

Over the last few years, I've found that my greatest obstacle to doing anything is that I don't feel like doing it Being mindful of that feeling has been a huge unlock in doing hard things for me.

I'm starting to think that the data-driven movement of the late 2010s/early 2020s was a mistake #databs, esp where A/B tests became the principal measure of success. It's a surefire way to make teams focus on shots on goal, not finding asymmetric value great for optimization, dangerous for culture

the jump from cron to airflow always baffled me. why isn't there something lightweight that improves on cron, but isn't so heavy as an orchestrator? I just want retries tbh cron is walking and orchestration is piloting a helicopter

Some things you have to experience firsthand - no amount of description or illustration will ever do them justice A good reminder that words and even images are lossy representations of underlying concepts

this is the real #databs I've been looking for

this is so helpful for anyone routinely switching btwn psql and more ergonomic warehouse dialects ... it also (partly) explains why I dislike writing postgres queries so much

There needs to be social media that is truly social, and [tentatively] bsky feels like it's captured that essence. Traditional algorithms (and so the posts) are so formulaically virality-optimized, but really I just want to talk to thoughtful people.

finally cut the cord on my planck 😍

something I enjoy about bluesky is that there seem to be a lot less people with motive and more people just having conversations

guy at Panera typing on cherry blues like he's summoning every cicada in a 5 mile radius

Is there a world where data people make more decisions? Esp as AI automates technical work, I wonder if giving an analyst context > giving a stakeholder raw access to a text-to-sql engine? We're strong thinkers, just missing context. wdyt #databs @machsci.bsky.social @imightbemary.bsky.social

Is there a world where data people make more decisions? Esp as AI automates technical work, I wonder if giving an analyst context > giving a stakeholder raw access to a text-to-sql engine? We're strong thinkers, just missing context. wdyt #databs @machsci.bsky.social @imightbemary.bsky.social

does anyone know any good libraries / resources for data anonymization? #databs

👍 also, this dynamic creates serious problems to some extent, ofc higher pay is a reasonable motivator as itll attract higher skill, but it can't make people care. you end up with an industry of mercenaries

Katie is spot on here data won't absolve you of personal responsibility when you can't make a decision some folks take "data-driven" way too far

my daughter just called a volcano a vanana today imo she's right though they should be called vananas

power's out and I'm running out of arm strength to read books to my 5yo by flashlight and kindle recs to tide us over until bedtime?

this should be a warning in front of every viz tool

do you primarily view your bi tool as a constructor for dashboards or visualizations? data products or one-offs? and why? #databs feels like there's some correlation here to tool choice, corporate culture, etc. (excel, eg, is obviously not a db creator, and best suited for a heavily ad hoc world)

so what's the best BI tool at this point? have we reached any sort of consensus? #databs

wanted to make a joke about another stakeholder asking ml/data people for a dashboard, but it's interesting that this is one of the few times a dashboard request is both valid and interesting a function of number of users, lack of incumbent dashboards, infrastructure/ml problems all around, prob

In general, I'm against layering models over models over models, but entropy as a concept for directing token prediction is super compelling. Really great overview on how entropix implements this by @kelloggt.bsky.social

I think a nontrivial part of why I'm on bluesky is just that it's a lot more performant than X

how AI in all technical domains will play out: - AI separates business logic from execution by handling and commoditizing execution logic - new logic-first abstractions arise that make specification of logic easier - jobs become about managing the new abstractions #databs

does anyone know if market changes are driven primarily by the sentiments of a select few, or sentiment of the masses? i.e. what's the power law exponent of the rank-ordered distribution?

Another reason why SQL is so compelling: it's good at separating logic from execution you shouldn't have to feel like editing logic will break execution, and the line is far too blurry in "proper" programming languages. used to think it was an interface problem, but it's more fundamental imo

#databs there's something about how tools like nvim seem philosophically aligned with local data tools like duckdb, datafusion, even dbt I imagine the world will increasingly move towards lightning fast, local-heavy, open, baremetal utilities over the coming years.

This is the clearest explanation I've read of the evolution from Hadoop -> motivation behind Iceberg from legend @julien.ledem.net. Great read

This is a fascinating take: when SQL/dashboard-generation become commoditized, analytics will center around formalism Can't hide behind a pile of trashboards anymore, will actually have to think

I want to state for the record that I still believe that analysts should become more strategic (esp now that I've exited my primary conflict of interest). Analysts need to become either more strategic or more technical, and this will be increasingly important as LLMs commoditize SQL-writing.

Something I've been thinking about lately: It's bizarre that AI and data are so closely related, but it's so ambiguous how data people are supposed to leverage AI. The best we have are mediocre text-to-SQL vendors and requests from execs to sprinkle some AI on things. Thoughts #databs?

I wonder if the etl jobs powering these dashboards are really old or really new

my god you're all here.

Hello Bluesky! 😁 I'm Robert, a founder, former data scientist, and physicist by training. I like to think about two things: (1) data systems and (2) systems of mind (think.ryi.me). Follow me if you want to hear my ramblings on either. 🙂

Hello! 😁