Centropy

[Centropy @Centropy3]

@NielsRogge Is this a joke?

[Centropy @Centropy3]

@RaminNasibov Get the fuck out of my timeline.

[Centropy @Centropy3]

@ADuke08812702 Was nimmt man da heutzutage um den Graphen zu extrahieren? https://t.co/EpWIsIBm2h , https://t.co/yofri3eiR7 , https://t.co/TEesWIR1Hn oder ein customized agent workflow?

[Centropy @Centropy3]

@robinebers Lighton in retrieval and ocr usecases

[Centropy @Centropy3]

@Die_Libertaeren R.I.P.

[Centropy @Centropy3]

@manishamishra24 YouTube https://t.co/BS8aaxQPG9

[Centropy @Centropy3]

@jeremyphoward Searxng for search. Firecrawl for parse. Camoufox for interactions

[Centropy @Centropy3]

@hyperlough @AliCologne Falls ihr eure lokalen Modelle ein wenig mit Hayek und co finetunen wollt... https://t.co/0jqu0Yo34L

[Centropy @Centropy3]

@hyperlough @AliCologne Qwen 3.6 35B ist ein Hammer. ~100 t/s auf ner A6000. Parallel mineru oder Chandra für PDFs und unstructured documents und der harness steht

[Centropy @Centropy3]

@hyperlough @AliCologne Exakt so! Und searxng als Suchmaschine und firecrawl als Parser und camoufox, wenn mit Webseiten interagiert werden muss

[Centropy @Centropy3]

@AxelKamann Nicht korrekt. Die sind break even bei den front tier modellen. Selbst wenn es drastisch teurer werden würde, cursor ultra oder cc. Hermes agent local mit dem richtigen harness und gemma4 31b oder dem moe oder qwen3.5 27b läuft für einen bruchteil der kosten auf sonnet niveau

[Centropy @Centropy3]

@Mayhem4Markets How much tps?

[Centropy @Centropy3]

@antoine_chaffin it would be nice if these benchmarks would cover entire pipelines, like hybrid bm25s+gtemoderncolbert as candidates generators with a reranker as second step etc...

[Centropy @Centropy3]

@antoine_chaffin Baseline bm25s leading the longembed bench? 😀

[Centropy @Centropy3]

@tom_doerr Why? Just use the searxng itself. This is a wrapper around searxng , you don't need it.

[Centropy @Centropy3]

@DeRonin_ https://t.co/BwhyzKZP4M is the originator

[Centropy @Centropy3]

@bilal_limi @cursor_ai Its getting outa control. I don't like it.

[Centropy @Centropy3]

in my opinion cursor is big time better than antigravity. If the advantage of a Google product is superior, I will use it despite my personal aversion against a copycat product. If it's on par or slightly better I will stay by the originator. Nothing wrong with copycat, I just discount the copycat.

[Centropy @Centropy3]

@dipeshdotdev @Star_Knight12 I really hate googles approach to copycat every good idea. Deepwiki is the originator and it's pretty good. I stay with deepwiki. Same for cursor vs antigravity, Google+, Google hangouts, Google Duo etc... The only service Google made something unique in recent time is notebooks

[Centropy @Centropy3]

@ericzakariasson Also from commands, I expect that everything defined in the command is executed. Skills work differently: they describe a collection of abilities, and when a skill is triggered, only the relevant abilities may be applied—not necessarily everything listed in the skill.

[Centropy @Centropy3]

@ericzakariasson Why? Please don't! Commands I trigger manually, skills are triggered by agent's. This is a totally different thing!