What do some actual data scientists say?
Long thread from Kareem Carr (twitter)
Skin in the game is worth more than sexy tools
You can be super smart, but unless people trust you as a person, they won’t trust your algorithms.
Empathize with the people you’re working with, they don’t know about AI and ML and Data Science and Statistics. Throwing precision, recall, accuracy, ROC, F1 at folks isn’t what’s going to win them over to your solution.
Them, “I need data science for this”.
You (most of the time), “no, you don’t” @Christopher Seaman
Customer: yeah we could use ML for that
Me: I think a case statement will work just fine @matsonj
GLMs are the answer, like, 80% of the time
- Logistic regression is not fucked (cookie scientist)
- Regression, Fire, and Dangerous Things (elevanth.org)
Always ask, “if the answer were <THIS>, how would that change what you do?” Analysis is not a substitute for indecision.
Writing a wrapper around someone else’s intellectual property (e.g., OpenAI) is not a defensible startup.
The lie of every MDS architecture diagram is that there is a single warehouse at the center of the business. I’ve almost never seen such an architecture. @mullinsms
Prediction: data warehouses like @SnowflakeDB will move up the stack with tech like WASM, sqlite and @duckdb into browser and edge nodes.
They’ll provide libraries that will make where your data lives irrelevant, enabling instant analytics. - @n8agrin
All models are wrong, but some are useful
