Shitposting and Memes. π rmoff.info
Yorkshire, UK
- Tweets 42,932
- Following 643
- Followers 9,960
- Likes 23,639
Pinned Tweet
Not really a "hot take", more like stating the bleedin' obvious: AI Slop is Killing Online Communities
rmoff.net/2026/05/06/ai-slopβ¦
4
8
1,156
This. Every word, of this.
@mipsytipsy - Confessions of an Unrepentant Slop Snob
charity.wtf/p/confessions-ofβ¦
1
108
Robin Moffatt π»ππ₯ retweeted
omg "you're right to push back" was just sitting there
you think Sisyphus had it bad but at least the boulder wasn't always saying "You're absolutely rightβand this one is my fault. Your instructions said not to roll down the hill, but that's exactly what I did."
7
38
1,114
28,838
well shit, we knew minio was dead, and here's another nail in the coffin.
at least it gives me chance to give a self-serving shill for an article i wrote earlier this year about MinIO alternatives ;) rmoff.net/2026/01/14/alternaβ¦
7
450
π New blog post from Gustavo de Morais:
Flink SQL Evolution: Handling Custom CDC with FROM_CHANGELOG and TO_CHANGELOG
developer.confluent.io/blog/β¦
9
476
My colleague Lorenzo Nicora has just published two excellent posts all about Flink watermarks, check them out!
1οΈβ£ Part 1: Apache Flink
developer.confluent.io/blog/β¦
2οΈβ£ Part 2: Confluent Cloud for Apache Flink
developer.confluent.io/blog/β¦
2
6
683
In case you missed it: August's set of Interesting Links in Data & AI ππ»
rmoff.net/2026/08/20/interesβ¦
Readers so far liking a pretty representative set of the kind of subjects covered :)
6
435
Robin Moffatt π»ππ₯ retweeted
Back when I wrote technical articles for the Confluent blog, I had this constant argument with the editorial team:
They hated that I used the same words to describe the same thing. Especially when it meant saying "low latency" 20 times in a 500 word post. They suggested I mix things up with "performant", "fast", "real time", etc.
My opinion is that using different words will be imprecise at best and confusing at worst. Using the same words to describe a concept whenever it comes up, helps readers know that I am still talking about the same thing and not introducing new ideas. For me, this was good writing.
Fast forward 6 years, and I'm using LLM that was clearly trained by the same editors who drove me nuts. And I think the harnesses also penalize models for repeating the same words in the output.
So every single time I need to ask: "You described this part of the design as 'retry path' and the other part as 'drive forward after failure' - do you mean retry in both cases? or is 'drive forward' somehow different from retry?
And no matter how often I say "please use the same terms when describing the same idea", it doesn't help - the LLM remains bound by its training and penalties, and I get to try and decypher what it means each time.
51
34
660
22,086
Here are the Interesting Links in the Data & AI world for August (slightly early this month in order to fit around my ποΈ holiday, since you ask). It's slightly shorter than usual, but the quality is still π₯.
π rmoff.net/2026/08/20/interesβ¦
2
346
Up & to the right! Is how the audience for my monthly newsletter "Interesting links in Data & AI" is going :) This month's will be shorter because of holidays (en-us: vacation), but there's still time for YOU to do two things:
(1) send me your submissions today! doesn't have to be your own article, could just be something interesting you read this month and
(2) subscribe to the RSS feed (rmoff.net/categories/interesβ¦) or the substack that I x-post to (interestinglinks.substack.coβ¦)
1
6
363
Warner Music Group - Why We Migrated From Cassandra to PostgreSQL RDS tech.wmg.com/rethinking-nosqβ¦
1
7
61
10,247
Gunnar Morling - A Fast Path for Fixed-Length Lists in Parquet developer.confluent.io/blog/β¦
3
7
1,030
Mark Callaghan - Orasort (Common prefix skipping, adaptive sort)
smalldatum.blogspot.com/2026β¦
4
387
