iiinova — where your digital self keeps evolving

Robert Yi @ryi.me · 05/03/2025

so much sass from 4o

040

Robert Yi @ryi.me · 14/02/2025

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

Robert Yi @ryi.me · 28/01/2025

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.

121

Robert Yi @ryi.me · 05/12/2024

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

1110

Robert Yi @ryi.me · 03/12/2024

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

010

Robert Yi @ryi.me · 03/12/2024

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

760

Robert Yi @ryi.me · 27/11/2024

imagine if the real modern data stack was oracle all along

140

Robert Yi @ryi.me · 27/11/2024

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.

230

Robert Yi @ryi.me · 25/11/2024

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

370

Robert Yi @ryi.me · 24/11/2024

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

7220

Robert Yi @ryi.me · 22/11/2024

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

130

Robert Yi @ryi.me · 22/11/2024

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

0141

Robert Yi @ryi.me · 22/11/2024

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

120

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.

170

finally cut the cord on my planck 😍

6170

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

0150

Robert Yi @ryi.me · 19/11/2024

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

040

Reposted by Robert Yi

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

9122

Robert Yi @ryi.me · 19/11/2024

9122

Robert Yi @ryi.me · 18/11/2024

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

261

Reposted by Robert Yi

👍 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

241

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

2130

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

150

Robert Yi @ryi.me · 16/11/2024

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?

030

this should be a warning in front of every viz tool

Robert Yi @ryi.me · 15/11/2024

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)

210

Robert Yi @ryi.me · 15/11/2024

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

17151

Robert Yi @ryi.me · 14/11/2024

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

110

Robert Yi @ryi.me · 13/11/2024

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

Robert Yi @ryi.me · 13/11/2024

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

030

Robert Yi @ryi.me · 13/11/2024

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

131

Robert Yi @ryi.me · 12/11/2024

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?

000

Robert Yi @ryi.me · 11/11/2024

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

030

Robert Yi @ryi.me · 11/11/2024

#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.

2100

Robert Yi @ryi.me · 10/11/2024

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

071

Robert Yi @ryi.me · 08/11/2024

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.

131

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?

9180

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

010

my god you're all here.

2132