Kirsten Lum
@machsci.bsky.social
🌲🌲 Applied ML/AI, data science, MLOps | Wife of 1, mom of 2 | Co-Founder and CTO of http://storytellers.ai
python 🐍 AI 🤖 cloud ☁️ data 📊
I also talk about Jesus here: @itskirstenlum.bsky.social
python 🐍 AI 🤖 cloud ☁️ data 📊
I also talk about Jesus here: @itskirstenlum.bsky.social
Pinned
Kirsten Lum
@machsci.bsky.social
· Oct 26
Hey bsky — thank you for all the follows.
I talk about machine learning, analytics, problems with data management (and what my team is doing to fix it), with a sprinkle of “why relational data is still the most important data.”
Looking to follow the curious and the courageous.
👋
I talk about machine learning, analytics, problems with data management (and what my team is doing to fix it), with a sprinkle of “why relational data is still the most important data.”
Looking to follow the curious and the courageous.
👋
Reposted by Kirsten Lum
Trying to make each and every transform efficient in the micro leads to inefficiencies in maintenance in the macro
January 27, 2025 at 9:46 PM
Trying to make each and every transform efficient in the micro leads to inefficiencies in maintenance in the macro
Startup idea — a toy that looks like a generic hand-held computer with a few analog inputs/outputs (buttons, switches, lights) and a screen. Kid can use voice to describe an adventure they want to have (exploring a jungle, space rescue mission, vet on Mars). Toy generates a relevant UI onscreen…
January 19, 2025 at 12:01 AM
Startup idea — a toy that looks like a generic hand-held computer with a few analog inputs/outputs (buttons, switches, lights) and a screen. Kid can use voice to describe an adventure they want to have (exploring a jungle, space rescue mission, vet on Mars). Toy generates a relevant UI onscreen…
Currently solving a thousand dynamic problems (watching Peppa Pig)
January 17, 2025 at 2:41 AM
Currently solving a thousand dynamic problems (watching Peppa Pig)
Life by 1000 kindnesses
January 16, 2025 at 5:20 AM
Life by 1000 kindnesses
By the way, the judo move if you’re in a conversation with someone like this is:
“Why won’t {solution} work?”
(They give an example)
“You mean you can’t figure out how to solve that problem?”
“Why won’t {solution} work?”
(They give an example)
“You mean you can’t figure out how to solve that problem?”
It’s always easiest to take the position that something is unsolvable, because it puts the burden on the other person to provide a solution.
But you can’t be wrong too many times before no one wants to work with you anymore.
But you can’t be wrong too many times before no one wants to work with you anymore.
January 15, 2025 at 7:27 PM
By the way, the judo move if you’re in a conversation with someone like this is:
“Why won’t {solution} work?”
(They give an example)
“You mean you can’t figure out how to solve that problem?”
“Why won’t {solution} work?”
(They give an example)
“You mean you can’t figure out how to solve that problem?”
It’s always easiest to take the position that something is unsolvable, because it puts the burden on the other person to provide a solution.
But you can’t be wrong too many times before no one wants to work with you anymore.
But you can’t be wrong too many times before no one wants to work with you anymore.
January 15, 2025 at 7:22 PM
It’s always easiest to take the position that something is unsolvable, because it puts the burden on the other person to provide a solution.
But you can’t be wrong too many times before no one wants to work with you anymore.
But you can’t be wrong too many times before no one wants to work with you anymore.
Woman culture is interesting and awesome.
January 8, 2025 at 4:26 PM
Woman culture is interesting and awesome.
One thing I like about this place is how folks list the company they work for in their bio. Look at us. We’re incredible. All of us with our lives and lil jobs
January 6, 2025 at 11:18 PM
One thing I like about this place is how folks list the company they work for in their bio. Look at us. We’re incredible. All of us with our lives and lil jobs
Me: *fills myself up with delicious, nutritious food*
My body: Thanks, now time for a little treat
My body: Thanks, now time for a little treat
January 6, 2025 at 8:11 PM
Me: *fills myself up with delicious, nutritious food*
My body: Thanks, now time for a little treat
My body: Thanks, now time for a little treat
Know your triggers, and have a plan for how to cope with their effects in a healthy way
January 6, 2025 at 4:43 AM
Know your triggers, and have a plan for how to cope with their effects in a healthy way
I love when Instacart suggests wines to pair with the string cheese I just added to my cart.
January 6, 2025 at 2:44 AM
I love when Instacart suggests wines to pair with the string cheese I just added to my cart.
Don’t think of it as whether the data has error or not. Most of the time it will.
Think of it as whether the error in the data will cause you to make a different decision than if the data was perfectly clean. Most of the time it won’t.
Think of it as whether the error in the data will cause you to make a different decision than if the data was perfectly clean. Most of the time it won’t.
January 2, 2025 at 4:49 PM
Don’t think of it as whether the data has error or not. Most of the time it will.
Think of it as whether the error in the data will cause you to make a different decision than if the data was perfectly clean. Most of the time it won’t.
Think of it as whether the error in the data will cause you to make a different decision than if the data was perfectly clean. Most of the time it won’t.
Reposted by Kirsten Lum
Conversely, database tools should offer some affordance for enforcing correct joins, especially in cases where identifiers are not unique in this way. But they don’t! Inexplicable to me.
December 31, 2024 at 6:39 PM
Conversely, database tools should offer some affordance for enforcing correct joins, especially in cases where identifiers are not unique in this way. But they don’t! Inexplicable to me.
Reposted by Kirsten Lum
And all the cloud platform maxxers were like
December 31, 2024 at 6:14 PM
And all the cloud platform maxxers were like
One of the biggest things I learned this year is there are several things you can do with your body to activate debug mode on your mind.
They’re so simple, the mind generally won’t put up much resistance to doing them.
But in order for it to work, you have a template for a healthy mind to install.
They’re so simple, the mind generally won’t put up much resistance to doing them.
But in order for it to work, you have a template for a healthy mind to install.
December 31, 2024 at 6:03 PM
One of the biggest things I learned this year is there are several things you can do with your body to activate debug mode on your mind.
They’re so simple, the mind generally won’t put up much resistance to doing them.
But in order for it to work, you have a template for a healthy mind to install.
They’re so simple, the mind generally won’t put up much resistance to doing them.
But in order for it to work, you have a template for a healthy mind to install.
Identifiers which are only unique within another identifier’s scope (e.g., product IDs that are only unique within their category) are the submarine screen doors of data.
December 31, 2024 at 5:56 PM
Identifiers which are only unique within another identifier’s scope (e.g., product IDs that are only unique within their category) are the submarine screen doors of data.
If you give an analyst tabular relational data, they’re going to build a world model, part 1
December 30, 2024 at 9:58 PM
If you give an analyst tabular relational data, they’re going to build a world model, part 1
3 year-old: This is a special coffee. It’s called — juice! Do you want to try some?
December 29, 2024 at 5:04 PM
3 year-old: This is a special coffee. It’s called — juice! Do you want to try some?
The #1 skill for delivering ML models at an eye watering pace is setting performance goals.
The goal is not to build the best model. The goal is to build a model that beats the status quo, and get it in production as fast as you can.
It works like this:
The goal is not to build the best model. The goal is to build a model that beats the status quo, and get it in production as fast as you can.
It works like this:
December 28, 2024 at 9:14 PM
The #1 skill for delivering ML models at an eye watering pace is setting performance goals.
The goal is not to build the best model. The goal is to build a model that beats the status quo, and get it in production as fast as you can.
It works like this:
The goal is not to build the best model. The goal is to build a model that beats the status quo, and get it in production as fast as you can.
It works like this:
Defining a sustainable business optimization metric that is not dollars is extremely hard.
December 27, 2024 at 9:55 PM
Defining a sustainable business optimization metric that is not dollars is extremely hard.
When something is going wrong, try skipping over the, “Why is this happening?” phase, and going straight to the, “What would happen if…” phase. Not only is the problem usually solved faster, you often learn the why in the process.
December 27, 2024 at 9:51 PM
When something is going wrong, try skipping over the, “Why is this happening?” phase, and going straight to the, “What would happen if…” phase. Not only is the problem usually solved faster, you often learn the why in the process.
AI doesn’t have a problem writing code. It has a problem creating a system of inputs and outputs.
December 27, 2024 at 8:34 PM
AI doesn’t have a problem writing code. It has a problem creating a system of inputs and outputs.
Claud’s favorite way to gaslight me: “Ah apologies, that code only works with newer versions of this package. Let’s try something else.”
December 27, 2024 at 7:29 PM
Claud’s favorite way to gaslight me: “Ah apologies, that code only works with newer versions of this package. Let’s try something else.”
Because all data is either a snapshot of reality or a proposal for how to categorize and describe it (aka facts and dimensions)
It's funny that nowadays any really serious hobby eventually leads to you using and mastering spreadsheets.
December 26, 2024 at 3:34 PM
Because all data is either a snapshot of reality or a proposal for how to categorize and describe it (aka facts and dimensions)
Reposted by Kirsten Lum
Merry Christmas to all those who don’t celebrate but would still welcome sincere wishes for their peace and flourishing.
December 25, 2024 at 10:38 PM
Merry Christmas to all those who don’t celebrate but would still welcome sincere wishes for their peace and flourishing.