“Reinvent for insight. Reuse for impact.”
https://endler.dev/2025/reinvent-the-wheel/
In search of causality, composition & the right abstraction.
CTO @ Fanstake.
Happily married, father of 2, or 4 counting furry friends.
Software, computer and electrical engineer.
Ham radio operator (KK6ZBK). I've been using computers since the C=64, ran my own BBS. I enjoy programming, and keep Scala, Nim, Typescript and Flutter easily accessible on my toolbelt.
Social moderate, fiscal conservative.
Also known as @boia01 on twitter.
https://twitter.com/boia01
“Reinvent for insight. Reuse for impact.”
https://endler.dev/2025/reinvent-the-wheel/
https://www.accountablealgorithm.com/post/data-judging-itself
ML evaluations are often seen as True North, but we should interpret them as data judging itself.
I wrote an article about how teams misinterpret the ML evaluation. I argue that we should be skeptical of the things we are trying to optimize, and brutally honest about how our metrics can mislead us. Sometimes the proxies in our data, models, and evaluations can lead us to believe that we're doing a great job, but only because we have our backs turned to the real problem.
Remember, once LLM hallucination is solved, the next "easy" problem in line is establishing veracity of the training data.
The way things are going, hallucinations may be the easier of the two.
https://www.accountablealgorithm.com/post/the-winner-s-curse
ML models tend to give their highest scores to the data whose value they overestimate the most.
The “winner’s curse” is something I wish I had formally learned about much earlier in my career. It haunts us all across machine learning and data science: ranking content, choosing experiment variants, running ad auctions, even the tuning and selection of models themselves. Whenever we use a measurement to help make a selection, we are biasing that measurement.
Scala 3.7.0 has been released! 🎉
âś… [stable] SIP-58: Named Tuples
âś… [stable] SIP-52: Binary APIs
đź‘€ [preview] SIP-62: For comprehension improvements
đź§Ş [experimental] SIP-61: Unroll
đź§Ş [experimental] SIP-68: Reference-able Package Objects
@paulisci *white* male voters
What we need is RICO laws, but for the executive branch.
Sabine Hossenfelder Asks If Science Is Dying. It’s Not. https://www.mcgill.ca/oss/article/critical-thinking-general-science/sabine-hossenfelder-asks-if-science-dying-its-not
No, Tom Friedman, Robotaxis Won't Save Democracy https://www.theindex.media/no-tom-friedman-robotaxis-wont-save-democracy/
What makes us think we can govern artificial intelligence if we can't even govern ourselves and petty dictators?
Artificial intelligence will erode social capital like no other technology before it.
We are already on the slippery slope.
What happens when we permanently turn to algorithms instead of friends, confidants, co-workers or family for guidance, answers, connection, and ... meaning??
Society's foundation -- which is already cracking under the pick-axe of greed-driven technology platforms -- will finally buckle.
Vous voulez savoir si une démocratie fonctionne bien ?
Ne cherchez pas à faire une moyenne de l'application des principes démocratiques.
Regardez plutôt comment est traité le plus petit de ses habitants.
La qualité d'un État de droit se mesure au respect des droits du plus faible ressortissant de cet État.
Despite everything else that’s going on, fossil fuels dropped below 50% of the US power mix for the first time last month
https://electrek.co/2025/04/04/first-month-on-record-fossil-fuels-drop-below-50-of-us-power-mix/
NEW: Judge Boasberg finds probable cause for contempt against the Trump administration: "The Constitution does not tolerate willful disobedience of judicial orders — especially by officials of a coordinate branch who have sworn an oath to uphold it." storage.courtlistener.com/recap/gov.us...
🏆 @rollingstone.com wins the Headline of the Day Award This is how it’s done. At least someone understands the assignment. www.rollingstone.com/politics/pol...
Why can't I "copy by reference" in spreadsheets? I want to CTRL-C a cell (or range) and paste it as a formula referencing its location (e.g. A2:B12).
Shouldn't this be an option under "Paste special" ?
@matejcerny You may also enjoy https://github.com/aboisvert/tuplelogy
As reliable as Google Drive is, when viewed as an external system, it's a single point of failure. That's why we backup all data that's stored there - Jakub shares our approach, accompanied by an OSS project!
https://softwaremill.com/how-we-approached-google-drive-backups-in-our-workspace-organization/
Couldn't make it to #JavaOne? Worry not! We'll live-stream the highlights to the Java YouTube channel, starting tomorrow at 1500 UTC (0800 PDT, 1600 CET) with the opening keynote and the #Java24 release.
https://www.youtube.com/watch?v=mk_2MIWxLI0
See you then!