Alex Strick van Linschoten

🎉 The @zenml_io LLMOps Database just crossed 1,000+ case studies! 17 months of curating real-world LLM production stories. Here's how the collection grew over time 📈 What started as a side project in July 2024 is now the largest open collection of production LLM case studies.

@wjayesh wait until you start playing around with notifications for watchOS... even more randomness!

Nobody told me that the notification system in iOS is so majorly fked. What a chaotic mess

On this Nowruz day amidst war, quiet rituals of survival in Tehran by @Najmeh_Tehran: "The sweetest things arrive at the most brutal moments. In Tehran, under bombardment, the craving comes almost instantly — after the blast, after the tremor, when someone reaches for something sugary. Jam appears. A piece of pastry is broken in half and shared." ft.com/content/a2b3dc…

KubeCon Amsterdam is next week and @zenml_io will be there. I won't be at this one, but my co-founder, Adam, and the amazing team will be holding it down at booth 1341 on March 23-26. If you're building ML pipelines or background agents (hint for a big announcement 😉) and want to nerd out about what production-grade actually looks like, go find him. Booth 1341. Drop by and say hi 👋.

Pigs don't just feel fear themselves. They also feel the fear of other pigs — and try to comfort them. In one study, pigs watching another pig in distress first showed fear themselves and then tried to comfort the distressed pig through snout-to-snout contact. Pigs who'd previously endured the same ordeal reacted even more intensely — they seemed to recognize what the distressed pig was going through. (Goumon & Špinka, 2016.) On factory farms, pigs trapped in gestation crates aren't just feeling afraid themselves. They're feeling the fear of all the other pigs around them. But the crate's iron bars stop them from even turning toward their fellow pigs — let alone comforting them with their snout.

@pvncher well, Claude Code compaction... Codex is considerably better...

Never seen a better image showing what compaction does to a model lol

📍 Iran | Lives lost, homes destroyed, and daily routines shattered. At a time when families should be preparing for Nowruz, many are instead mourning loved ones. Vincent Cassard, ICRC head of delegation in Tehran, shares. Read more 👉🏽 ms.spr.ly/6013QorMX

I've been building panlabel — a fast Rust CLI that converts between dataset annotation formats — and I'm a few releases behind on sharing updates. Here's a quick catch-up. v0.3.0 added Hugging Face ImageFolder support, including remote Hub import via --hf-repo. You can point it at a HF dataset repo and it figures out the layout (metadata.jsonl, parquet shards, even zip-style splits that contain YOLO or COCO inside). v0.4.0 overhauled auto-detection so it gives you concrete evidence when format detection is ambiguous ("found YOLO labels/ but missing images/") instead of a generic error. Also added Docker images. v0.5.0 brought split-aware YOLO reading for Roboflow/Ultralytics Hub exports and conversion report explainability — every adapter now explains its deterministic policies so you know exactly what happens to your data. v0.6.0 is the big one. Five new format adapters: → LabelMe JSON (per-image, with polygon-to-bbox envelope) → Apple CreateML JSON (center-based coords) → KITTI (autonomous driving standard — 15 fields per line) → VGG Image Annotator (VIA) JSON → RetinaNet Keras CSV That brings panlabel to 13 supported formats with full read, write, and auto-detection. Also in v0.6.0: YOLO confidence token support, dry-run mode for previewing conversions, and content-based CSV detection. Single binary, no Python dependencies. Install via pip, brew, cargo, or grab a pre-built binary from GitHub releases. This is the kind of project I enjoy just steadily plodding away at — ticking off one format at a time until every common object detection annotation format is covered. Still sticking with detection bboxes for now, but the format list keeps growing. #ObjectDetection #Rust #MachineLearning #ComputerVision #OpenSource

This evening ends with some darkness. "Just for fun". dark.ronacher.eu/2026/3/14/the-…

Here's what a few hours in the life of Lebanon's National Lifeline for Suicide Prevention (1564) sound like during #war time: - A 23-year-old displaced mother from South Lebanon (Khyem) calling from inside a car in Saida (now known as the transit city) - where she is living with her 3 young children because shelters are full. - An 88-year-old man displaced from Southern #Beirut calling to ask for basic assistance - A young man calling from Southern Beirut, currently under attack, calling because he and his mother had nowhere safe to go. - Another caller described feeling “مخنوق” — suffocated by the density, the noise, the uncertainty created by war. And in the middle of all of this: - A 16-year-old girl who attempted suicide - A 23-year-old struggling with addiction after a suicide attempt yesterday - A mother with cancer caring for a child with epilepsy while seeking help for her suicidal daughter - A man who was about to end his life with shattered glass before picking up the phone This is the daily reality of answering Lebanon’s National Lifeline (1564), where every call reflects the collision of our crises: war, displacement, poverty, chronic illness, addiction, and the silent toll on mental health. In contexts like #Lebanon, a crisis hotline is much more than an emotional support line. It becomes a first point of access to the entire health and social protection system. In a single shift today, our team provided: - suicide intervention and safety planning - emotional support during acute distress - referrals for psychiatric care and medication - shelter and humanitarian assistance referrals - support for displaced families - guidance for parents responding to suicide attempts - connection to emergency services and NGOs Behind each call is a trained responder, a volunteer, holding space for someone who may have nowhere else to turn, while silently living many of the same struggles themselves. In times of war, mental health emergencies do not pause. They multiply, and set the stage for the long term psychological consequences that follow conflict. Lebanon’s National Lifeline continues to operate 24/7, responding to hundreds of calls each month from people facing the most difficult moments of their lives. Every call our volunteers answer, is a proof of why this Lifeline is essential to our national infrastructure, and must be protected and sustained.

@wjayesh you just missed @htahir111 !

so pumped that i'll be in sf 🌉 starting next week! who should i meet? 🙋‍♂️ also, any hackerhouses/meetups/builder spaces that i should def not miss?

Repo: github.com/strickvl/panla… Release notes: github.com/strickvl/panla…

@JunaidQadirB yeah when you build it yourself you realise it isn't actually SO crazy what's going on. you also realise that the devil is really in the details about all the different pieces. great exercise what you did!

Built barebones-ai to understand how agentic AI tools actually work. Core loop: prompt -> memeory -> tools -> loop ->response Built in PHP to learn mechanics by implementing. I know not the best language for this. Repo: github.com/yourusername/b…

@addyosmani @altryne @swyx @JPoehnelt write up was great, but what's this at the bottom 🤔

@altryne @swyx @JPoehnelt Justin is absolutely worth a follow and has a good write-up about this project too! justin.poehnelt.com/posts/rewrite-…

Introducing the Google Workspace CLI: github.com/googleworkspac… - built for humans and agents. Google Drive, Gmail, Calendar, and every Workspace API. 40+ agent skills included.

Kanban for “targeted workflows” 😔

"The targets you hit on the first day are the most well-planned, well-thought-through targets," she says. "This is when they should have had the maximum amount of intelligence .... It should have been their best, most precise day of the whole war. And they hit a school."

@pvncher Depends on the task tbh but for the vast majority of things I want the strongest and most robust option

Would you prefer a model that does what you wanted 90% of the time, or 98%? In long running execution any drift compounds so it means everything. Not to mention the importance of alignment and prompt injection risk.

@AndyJScott @corbtt x.com/corbtt/status/…

@corbtt @strickvl how is it accessing browser?

GPT-4 era browser automation was interesting but functionally useless. With Opus 4.5 it finally got reliable enough to be useful for many background tasks. With GPT-5.4 (fast mode, low thinking) it's now almost as fast as I am.

@pvncher @iannuttall yeah +1. I finally bit the bullet and followed the wizard's recommendations and had a glorious run today.

@iannuttall Interesting. I mainly use it in RP but have no such issues

anybody else felt like gpt 5.4 has been a bit "off" today? just feel like I am fighting with it on things it had no issue with before like building my chrome extensions for local vs prod, working with conductor ports, etc