Oliver Trosien (inactive)

We are hiring a Java Engineer at @ZalandoTech working on Telemetry/SRE systems in Berlin: jobs.zalando.com/en/jobs/463017… #observability #opentelemetry #monitoring #metrics

good morning and welcome to 2023

Ernie und Bert werden 50 - so sehen sie heute aus

Freier Eintritt ins Miniatur Wunderland für alle, die es sich ansonsten nicht leisten können an ausgewählten Terminen im Januar. Bitte helft uns, möglichst viele Menschen zu erreichen! miwula.de/leisten #wunderland #miwula

@MelMitchell1 instead of just prompting for a simple solution have you tried asking gpt3 for an explanation of their answer / the identified pattern? I'd argue that 'alphabetical successor' may be the most obvious one to most of us but there could be many others.

One of my new years resolutions is to blog (from time to time) about interesting work in AI. I'm trying out Substack for this. My first post is a perspective on recent paper by Webb et al., "Emergent Analogical Reasoning in Large Language Models". aiguide.substack.com/p/on-analogy-m…

sd: your script directory github.com/ianthehenry/sd

Our Tian Su, VP Customers for Profile & Personalization, talks on our podcast #ZalandoPiecesofMe about how to use data to give customers the personalized content & products they'll ♥️ More here: medium.com/zalando-talent…

Look how much I learned on Duolingo in 2022! How did you do? #Duolingo365

"We have normality. I repeat, we have normality. Anything you still can't cope with is therefore your own problem" - #DouglasAdams

Trombone Champ is the best game ever made

Am Ende dieser nervenaufreibenden Recherche kam es, wie es kommen musste: Eine Person im Gazprom-Netzwerk, Sigmar Gabriel, drohte mit rechtlichen Schritten gegen unsere Recherche. Natürlich haben wir sie heute trotzdem veröffentlicht - ein Mustread 🧵 correctiv.org/aktuelles/russ…

Bundesminister #Habeck zu #NordStream2.

#ErneuerbareEnergien sind zentral für mehr #Klimaschutz und #Energieunabhängigkeit. Aktuell entsteht eine neue Dynamik, die 🇩🇪 aus der Krise helfen wird. Bundesminister Robert #Habeck schildert die bisherigen Schritte. ➡️bmwk.de/Redaktion/DE/P… #Gaskrise #Gasumlage

10 minutes each day for a good habit, keeps the flame alive. #duolingo

No, planting a trillion new trees can NOT save the world. If we want a livable planet, and I would strongly suggest that we do want a livable planet, it will take ending the fossil fuel industry and toxic capitalism nytimes.com/2022/07/13/mag…

In some cases, an HTTP #API only uses POST because there's no better fitting alternative: POST is the "do anything" fallback of HTTP. But if you're using POST for GET-like requests just because GET doesn't allow request bodies, then the proposed QUERY method is a better choice.

#FreeBritney Remember when we covered this Britney classic with @HarmAndEase?

@saareliad the test ran on 6 shards to closely reflect prod (so eg. 80ms total extra latency if we add vector search). We are sharding to parallelize enough so that we don't run into timeouts even for our largest possible result set.

@otrosien @xeraa @jobergum Can you elaborate a little about the target tail latency and shards? e.g., does this mean that 80ms latency in the diagram benchmarked for 1 CPU node is approx. 80/6 ms when split to 6 shards? Is there some target latency after which you start sharding?

100+ nodes for this? I got questions

I really enjoy this monthly themed reward system. Very cute! Happy Anniversary #10yearsduolingo @duolingo

@xeraa @jobergum And finally, it's the backbone of the shop. If that system fails we're in big trouble. I sometimes call this way of over-scaling "insurance cost" - and I don't think we're the only ones doing it :) 6/6

@xeraa @jobergum It's not so bad as it sounds because the cluster auto-scales during the day between 50 and 120 nodes. We wrote our own Kubernetes operator for this. When scaling it is optimizing for balanced shard-to-node distribution. github.com/zalando-incuba… 5/6