Jim Dowling

At university, there was a maxim that, you could only do 2 of these 3:
study, sleep, party.

As a dad, I experience that I can only ever focus on 2 out of these 3 at any given point in time:
career, good parent, regular exercise.

Is it just me?

Artist's impression of the exoplanet K2-18b. Credit: A. Smith, University of Cambridge.

𝗡𝗼, 𝗮 𝗯𝗶𝗼𝘀𝗶𝗴𝗻𝗮𝘁𝘂𝗿𝗲 𝘄𝗮𝘀 𝗻𝗼𝘁 𝗷𝘂𝘀𝘁 𝗱𝗲𝘁𝗲𝗰𝘁𝗲𝗱 𝗶𝗻 𝗞𝟮-𝟭𝟴𝗯'𝘀 𝗮𝘁𝗺𝗼𝘀𝗽𝗵𝗲𝗿𝗲.

K2-18b is back in the news, now with a bold claim that biosignature molecules (DMS and/or DMDS) have been 'detected at 3σ'.

Most exoplanet astronomers are extremely sceptical about these claims, let's see why (1/n).

🔭🧪🪐 #exoplanet

6/10. "Build TikTok's Personalized Real-Time Recommendation System in Python with Hopsworks" ⸱ +3k views ⸱ 07 Oct 2024 ⸱ https://youtube.com/watch?v=skZ1HcF7AsM

YouTube video by Hopsworks Show, don't tell - Building Data Pipelines with AI

This is very cool. Lex and Aleksey show you how you can use an #LLM #assistant to generate code that automatically imports features into the Hopsworks feature store. This makes getting started with the #ai #lakehouse even easier! See youtu.be/wtEPONdKV6Y for another great episode of #showdonttell.

I was pleasantly surprised to see my "From zero to a working ML system with Python, free serverless services + FTI pipelines" talk ranked at 43 in most popular Python talks of 2024.
Talk
https://www.rfr.bz/be18657

Reference:
https://www.rfr.bz/b78475a

I was pleasantly surprised to see my "build Tiktok with Hopsworks" talk ranked at number 6 in most popular Python talks of last year.

TikTok with Hopsworks talk:
https://www.rfr.bz/beb40a4

Source: https://www.rfr.bz/b58c8ae

Foundation Model for Personalized Recommendation By Ko-Jen Hsiao, Yesu Feng and Sudarshan Lamkhede

This is an interesting dataset and paper by Netflix that in essence says:
"By predicting the next interaction, our foundation model must build an internal model of consumer behavior."

https://www.rfr.bz/b30d8e7

The top 10 fallacies of MLOps

www.hopsworks.ai/post/the-10-...

What tables in the Data Warehouse should be exposed to data scientists to create features from?
The gold layer is aggregated features. If you want to create your own aggregations, you need access to the raw events (bronze or silver layers).
Reference:
https://www.rfr.bz/b793ee3

Hopsworks near the top of hackernews today with an article on how we migrated our serverless platform from AWS to OVH (european cloud company).

The 10 Fallacies of MLOps - Hopsworks Discover the top 10 MLOps fallacies that hinder AI deployment. Learn how to avoid common mistakes, streamline workflows, and build scalable AI systems.

The top 10 fallacies of MLOps

www.hopsworks.ai/post/the-10-...

The 10 Fallacies of MLOps - Hopsworks Discover the top 10 MLOps fallacies that hinder AI deployment. Learn how to avoid common mistakes, streamline workflows, and build scalable AI systems.

The top 10 fallacies of MLOps

www.hopsworks.ai/post/the-10-...

JD Vance attacks Europe over free speech and migration The vice-president's speech at the Munich Security Conference was denounced by several European leaders.

There is a concerted effort to drive a wedge between Europe and the US by the new US Administration.

Apparently we should not be worried about Russia/Ukraine, but free speech and Greta are our big problems. I don't see Europe taking this lying down.

www.bbc.com/news/article...

Sweden has the highest number of Unicorns in the world outside of Silicon Valley Breakfast Business with Joe Lynam · Episode

I was on irish radio, Newstalk , this morning, talking with Joe Lynam about Hopsworks and AI in the EU.

open.spotify.com/episode/0yBg...

Congrats on taking the entrepreneurial plunge

Musk's nazi salute was deliberate. He's outflanking Steve Bannon to appeal to the Maga masses. He's afraid he'll lose access to his cash cow, Trump.

Tesla's proprietary charging infrastructure is now their weak point in Sweden. Massive queues. Unions have stymied rolling out Tesla charging stations. Commodity charging is fine.
As Peter Thiel said "Competition Is for Losers". This is the playbook for how to fight back.

x.com/i/status/187...

Not so subtle dig here from the BBC
"His performance came 12 years later than planned, after he'd originally been forced to pull out through injury after cutting his finger making a sandwich."

I was curious who Andy referred to as screwing over OtterTune. It's ASCII art in the HTML of their announcement that they were winding up.

Depends on what parts of Israel you include.

Good reproducible benchmark study by @coiled.io. I get this question all the time by MLOps folks. What size of workloads can frameworks handle :
* Pandas ~1 GB (this one i added for reference, not in the study)
* Polars ~10 GB
* DuckDB ~100GB to 1TB
* Dask and PySpark up to 10 TB

lnkd.in/dAxUAWC4

I went to fork the repo, but it's name just turned me right off.
Who calls a github repo 'all'?

Feature Importance - with LLMs performs as good as Lasso, mutual information, etc

arxiv.org/html/2407.02...

Mistype. I meant add zk to etcd, and you cover the set of all such datastores that need an external db (typically for correctness during failure or recovery)

You could easily replace etcd with zookeeper and you would cover nearly all such systems :)

I am surprised Confluent haven't taken down their posts. Buy and bury.

Breaking Down FSI Regulations for AI in 2025 - Hopsworks Hopsworks' AI Lakehouse capabilities, such as high availability, MLOps platform and Feature Store, can be useful for FSI's navigating regulations in 2025.

Looking forward to 2025, I have been working together with the #Hopsworks team to try to understand what the upcoming regulations are going to mean for the Financial Services Industry ( #fsi ). www.hopsworks.ai/post/breakin...

I approve this message :)

Correctness at Feldera In this blog post, we briefly describe our efforts and development processes that ensure Feldera's engine is correct.

We have a strong background in formal methods and system reliability. So we’ve been surgically applying these techniques to scale our correctness efforts at Feldera. We’ve found them quite effective at catching bugs early during development, long before they can manifest in production.

Databricks and Snowflake went high (vendor lockin at the catalog layer), AWS went low (at the S3 layer).
It is an abuse of their monopoly position, but they know that in the current regulatory environment nobody will do anything about it.