Brian Roepke

From DynamoDB to @AWScloud Lambda, learn how to build a powerful, scalable API. Follow @broepke's step-by-step guide to create and deploy your first serverless API and test it with a custom front end. towardsdatascience.com/build-your-fir…

Looking for a fun project walkthrough for inspiration? @broepke is here to help, with a clear, step-by-step recap of how he built a sudoku solver with OCR and a sprinkling of some ML magic. towardsdatascience.com/lets-learn-a-l…

Collection of insane and fun facts about SQLite. Let's go! SQLite is the most deployed and most used database. There are over one trillion (1000000000000 or a million million) SQLite databases in active use. It is maintained by three people. They don't allow outside contributions.

@svpino I didn’t visit help documentation in 2024

I did not visit StackOverflow in 2024.

@Scobleizer Welcome back? :)

I have muted everyone in tech. I am unmuting now that I know that it hurts your reach. But only if you are following me and you leave a comment here. That way I know you haven't muted me. I will leave the rest muted because Elon said that if I bug those who have muted me, I'll be marked as a spammer and penalized. Everyone is on my lists: x.com/scobleizer/lis…

Bringing together programming and computer vision techniques, @broepke walks us through the entire workflow of creating a sudoku solver. towardsdatascience.com/lets-learn-a-l…

I love @PrefectIO - The upgraded to 3.0 and gained a ton of efficiency in my AWS infrastructure!

For fun, I wrote a Wordle Solver. It's a common coding problem attempted many times. While this is not new, I'm pleasantly surprised by how well it performs. 30 consecutive games from the archives without a miss and most solves within three guesses! hubs.li/Q02ClmHf0

@tunguz So you’re saying deep learning for tabular data? 😂

Not getting personally attached to your mental models is a superpower. Don’t be too trigger happy to change them, but don’t hesitate to do so when the overwhelming new information warrants.

The first time I heard about Groq, I dismissed it. But this team is legit! Their performance for LLM Inference is unbelievable! LPUs, or Language Processing Units, are a new type of chip architecture that happens to be the perfect use case for it! hubs.ly/Q02ny3bz0

This is such an important rule for building a high-performance team. Performance Vs. Trust by @simonsinek hubs.ly/Q02ntKC20

People who don’t use AI are ngmi, and people who use AI are being used by AI for RLHF and are also ngmi.

We did it. #TwinsWin! #WeBelieveinTC

@tunguz “Verbal kint”

The greatest trick the AGI ever pulled was convincing the world it wasn’t already here.

@tunguz You missed cold plunge

Can confirm.

@svpino I got lucky when I first started I stumbled upon this advice and never looked back. It’s 10x easier as well than trying to do each step independently.

Start using pipelines. It's the simplest way to 10x your Machine Learning setup. The idea of pipelines has been around for a long time, yet many people ignore them or think they are only helpful in making your code more readable. They are much more than that. A pipeline is an independent sequence of steps organized to automate a process. One of the main advantages of using one is the ability to reuse the process at different stages and with different datasets. You should create a pipeline to transform your dataset at the beginning of a project. You can reuse the same pipeline to transform production data before running it through the model. The absence of a pipeline is an immediate red flag and a sign that you are, at best, duplicating code or, worse, transforming production data in a different way than you did the training dataset. Attached you'll see an example pipeline for a straightforward problem. Pipelines aren't just nice to have; they are a fundamental requirement to build Machine Learning systems.

A penny doubled each day for a month: day 1: $0.01 day 2: $0.02 day 3: $0.04 day 4: $0.08 day 5: $0.16 day 6: $0.32 day 7: $0.64 day 8: $1.28 day 9: $2.56 day 10: $5.12 day 11: $10.24 day 12: $20.48 day 13: $40.96 day 14: $81.92 day 15: $163.84 day 16: $327.68 day 17: $655.36 day 18: $1,310.72 day 19: $2,621.44 day 20: $5,242.88 day 21: $10,485.76 day 22: $20,971.52 day 23: $41,943.04 day 24: $83,886.08 day 25: $167,772.16 day 26: $335,544.32 day 27: $671,088.64 day 28: $1,342,177.28 day 29: $2,684,354.56 day 30: $5,368,709.12 day 31: $10,737,418.23

@tunguz @marktenenholtz The ever speaker of simplicity and truth.

@marktenenholtz tf-idf + LR is probability good enough for most such use cases actually.

If you're new to LLMs, you'll probably be surprised at the sheer effectiveness of ~150-300M parameter encoder-only models. If you have a classification task and some training data, you're probably wrong not to start there.

@wholemars 🤣🤣🤣

So long to the Blue Bird icon on my home screen… gone but not forgotten 😢

@svpino 💯💯💯

A must for every programmer: • Basic data structures • Sorting and searching • Graphs and trees • Approximation algorithms • Dynamic programming • Complexity analysis This list is neither exhaustive nor optional, despite what many anons will tell me in the replies to this post. Of course, you can make money without knowing any of this. You can have a career and live happily ever after, but how much will you leave on the table? Somebody always wonders how knowing Dijkstra's algorithm helps write CSS and HTML. The answer is simple: it's your way out of writing CSS and HTML for the rest of your life! And now we have Artificial Intelligence. And every report I read says that coding will be one of the more vulnerable areas. It makes sense: coding is very structured, easy to verify, and most of us hate writing code anyway. Automating "writing the code" makes a lot of sense. But deciding what code to write in the first place is the hard part. Programmers aren't going anywhere, but that doesn't mean everyone is safe. A thought exercise for you: Who will be the last standing? Is outsourcing your knowledge to third-party libraries a good, long-term strategy? Yesterday, I watched a video of Daviz Muñoz, a 3 Michelin star chef recognized as 2022's Best Chef in the World. He mentioned that the better he got at cooking, the more time he spent studying. Day and night. Studying non-stop. Think about that.