Claude & ChatGPT 4.5 Drop, A Robot Goes Full Kung Fu, and AI Chatbots Start Speaking in Beeps
Catch up on eBay’s in-house LLM, Strava’s 20x faster deployments, and Datadog’s game-changing architecture overhaul.
Hey Fellow Tinkerers!
Thank you to everyone who voted last week about the new format. Considering the response (100% Yes), I will continue the weekly update. So moving forward you will receive the following:
1 article summarizing the main data news (the one below)
1 article covering a topic in either data science, engineering or analysis
Always looking forward to your feedback. You can share it by replying to this email or leaving comments below. With that out of the way, let’s get to the main bit
The Buzz 🐝
Both Anthropic and OpenAI released their latest models last week. The early reviews of ChatGPT 4.5 are not very positive, specially taking into account its cost vs its performance:
A humanoid robot at a Chinese tech festival allegedly went full kung fu mode on attendees. Officials suspect a software glitch was at play.
Two developers created a custom protocol called Gibber Link that enables AI agents to communicate more efficiently by switching from human language to a sound-based protocol, reducing communication time by 80% and lowering computational costs by 90%. (Check the video below to hear them “talk”)
Data Science & AI
- provides a good overview of Multimodal LLMs capable of processing different data types such as text, images, and audio.
Scaling Large Language Models for e-Commerce
How eBay developed e-Llama, an in-house LLM by adapting Meta's Llama 3.1 to achieve 25% improvement in e-commerce-specific benchmarks
Cracking the ETA Code: How Lyft Improves its Predictions
How did Lyft improve its Estimated Time of Arrival (ETA) calculation to minimize rider frustration
Data & Analytics Engineering
Data Pipelines in Machine Learning Systems
provides a good hands-on tutorial catered to data ingestion for ML systems
Rain: A Key-Value Store for Strava’s Scale
Learn How Strava’s ‘Rain’ Key-Value store accelerated deployment times from 20 minutes to 1 minute
How Datadog Achieved 99% Timeout Reduction with 20x Scalability Boost
Discover the architecture that cut costs by 50% and unlocked massive scalability for Datadog
Data Analysis and Vizualisation
The Base Rate Fallacy: When Ignoring the Big Picture Leads to Bad Decisions
Learn how focusing on specifics can trick even the smartest analysts
Charted: The Pyramid of S&P 500 Returns
Great visualization of S&P returns going back to 1874 by Visual Capitalist
Happy Ending
“Data Engineering” in a nutshell
That’s so wild. The machine doesn’t look like it was responding to the person. It looks like it’s responding to the barrier.