@rotecodefraktion.bsky.social
Das @zdfde.bsky.social hat kein Problem damit Neunazis groß in Szene zu setzen, aber wenn zum Widerstand gegen die Faschisten musikalisch aufgerufen wird dann ist Zensur kein Problem. Sagt alles über die Intendanz und Geschäftsleitung des ZDF. #nazis #steigbügelhalter
#Ornith 1.0, a coding model that learns its own agent scaffold. Tested locally on the M3 Max, GGUF against #MLX, with a surprising stumbling block in the packaging. → www.rotecodefraktion.de/en/blog/orni...
Ornith 1.0: When the Model Writes Its Own Scaffold
DeepReinforce trains a coding model that learns its own agent harness. I tested the 9B and 35B locally on an M3 Max. The idea is strong, the benchmarks are self-reported, and the biggest stumbling blo...
rotecodefraktion.de
A bigger context window isn't a memory. New piece on LLM memory for AI agents: CoALA's four functions, the real cost of RAG→GraphRAG, and memory poisoning (the failure mode nobody names). Plus what n8n does natively. www.rotecodefraktion.de/en/blog/n8n-... #n8n #LLM
What an Agent Remembers: LLM Memory between Context Window and Knowledge Graph
Agent memory is an architecture problem, not a toggle. An ordered overview: the cognitive memory functions, why the context window is not memory, RAG through GraphRAG with their real costs, and where ...
rotecodefraktion.de
You know that, one is tired but still wants to finish the last shot. To cheer up, I ask the Beta Siri to make the tip of my nose a little sharper. I don’t know...
Why I don‘t trust #ChatGPT & #Microslop for AI Agents? Quite simple, they are forcing their customers into loops like „Do you want this and that? Give me a go…“ and at the end of the day, one spends tokens for bullshitting. And this is not a problem of the LLM; it’s broken by design. SGFY #AI
Local Python showdown on an M3 Max ( #MLX, real execution): #Gemma 4 12B vs a community coder vs Qwen3-Coder-30B. On standard tasks all tie; #Qwen wins on speed (~92 tok/s — MoE, ~3B active). The arithmetic parser beat all three www.rotecodefraktion.de/en/blog/gemm...
Local Python on the Mac: Gemma 4, Its Coding Finetune, and a Current Specialist
Gemma 4 12B is celebrated as the most complete local model. For Python I ran three variants against each other on an M3 Max, with real code execution, including a parser that defeats all three.
rotecodefraktion.de
Finale of the n8n series: n8n authenticates to SAP BTP via client-credentials flow against xsuaa, then calls a CAP service, CDS, OData, xsuaa-secured, like a real S/4 target. And one service-key field must never hit the repo. www.rotecodefraktion.de/en/blog/n8n-... #n8n #SAP #BTP #OAuth #AI
n8n Authenticates Against SAP BTP — OAuth Against a CAP Service
The series finale: n8n binds to its own service on SAP BTP with a real OAuth flow. Client Credentials against the xsuaa token endpoint, a CAP service as a SAP-native stand-in, the n8n OAuth2 credentia...
rotecodefraktion.de
Multi-Token Prediction for MLX, #MTPLX tested on the M3 Max. 2.24× promised, +40% measured on my M3 Max (same quality). www.rotecodefraktion.de/en/blog/mlx-... #MLX #AppleSilicon #AI
Multi-Token Prediction for MLX: MTPLX Tested on the M3 Max
MTPLX promises noticeably faster local LLMs on the Mac via Qwen 3.6's built-in MTP heads. Measured on an M3 Max, with real numbers instead of marketing claims.
rotecodefraktion.de
#QuEra 's counter-bet to Microsoft's Majorana chip: 256+ logical qubits by 2028 on AWS Braket, w/o exotic physics, just ordinary atoms & error correction. qLDPC: ~2 physical qubits per logical one, not thousands. #QuantumComputing www.rotecodefraktion.de/en/blog/quera-libra-quantencomputer-2028/
QuEra Libra: The Counter-Bet to Microsoft's Quantum Chip
QuEra wants to ship a fault-tolerant quantum computer with more than 256 logical qubits in 2028. The lever is not exotic physics but error correction with unusually little overhead. An assessment with...
rotecodefraktion.de
Microsoft's quantum chip computes with a particle that might not exist. #Majorana 2's long parity lifetime proves neither a working qubit nor that Majoranas are real — a trivial Andreev state fits too. Microslop again? #science #quantumchip www.rotecodefraktion.de/en/blog/majo...
Majorana 2: Microsoft's Quantum Bet on an Unproven Particle
Microsoft celebrates Majorana 2 as a 1,000 times more stable quantum chip. Its foundation is Majorana quasiparticles, whose existence physics has not settled to this day. An assessment with technical ...
rotecodefraktion.de
Has anyone applied to @anthropic.com on behalf of Europe yet? No? Then I did. Freedom, rule of law, top talent that won't get deported tomorrow — and militaries that respect your model's red lines. www.linkedin.com/pulse/applic... #claude #fable5 #mythos5
Application for Europe
From: Europe (represented by Germany, Department of "We've Got an Idea, Actually" and on the ground by me David Krcek ) To: Anthropic, San Francisco Re: Unsolicited application for the role of locatio...
linkedin.com
Article 11 of the series "Getting Started with n8n": n8n in Queue Mode — Main, Worker, Redis and Python. Main accepts and enqueues, workers execute. When it is worth it, how workers scale, and why queue mode is not retry. www.rotecodefraktion.de/en/blog/n8n-... #n8n #QueueMode #Redis #Python
n8n in Queue Mode — Main, Worker and Redis
Operation goes production-ready. This article moves the single-process instance to queue mode: main accepts webhooks and enqueues executions in Redis, workers run them. With worker scaling, queue metr...
rotecodefraktion.de
The same CSS function, two colors — depending on dark mode. Sass resolves at build time, native @function at render time. That's the new part, not the catch-up. #CSS #ModernCSS #Frontend #WebDev #WebDevelopment With live demos: www.rotecodefraktion.de/en/blog/css-...
CSS Gets Real Functions: The @function Rule
With @function you define your own CSS functions. The real novelty is not that CSS rebuilds Sass, but that a CSS function decides in the browser instead of at build time.
rotecodefraktion.de
Addendum 06/14/26: This is turning into a business thriller with the Trump regime implicated. AWS — Anthropic's biggest investor AND infrastructure provider — flagged the models to David Sacks, ex-AI czar and #Anthropic critic. www.rotecodefraktion.de/en/blog/us-r...
Three Days at the Frontier: The Trump Regime Pulls Fable 5 and Mythos 5
Three days after launch, an export-control directive forces Anthropic to disable Fable 5 and Mythos 5 for all customers. What happened, why, and what it means.
rotecodefraktion.de
Cheapest reliability upgrade for any n8n setup: a two node error workflow that pings you the second something breaks. Telegram, Slack, email, whatever. Built it faster than I wrote this. Setup: floxolab.com/blog/n8n-err...
Protect n8n Workflows with Chat Error Alerts | FloxoLab
A simple 2-node n8n Error Workflow can alert you in Telegram, Slack, Google Chat, email, or another channel when business automations fail silently.
floxolab.com
Container machine, Apple's quiet attack on #Docker Desktop. A dedicated micro-VM per container instead of Docker's shared VM. The architecture, the memory behavior under load, and why Apple is really going after Docker Desktop. #AppleSilicon #macOS #WWDC26 www.rotecodefraktion.de/en/blog/appl...
Container machine: Apple's Quiet Shot at Docker Desktop
With the Containerization architecture and the new Container machine, Apple has spent two years building a native Docker alternative for the Mac. A technical analysis of who is really under attack.
rotecodefraktion.de
ChatGPT sounds thoughtful. Claude sounds empathetic. And yet there's no one home. Why confusing a language model with consciousness isn't a misunderstanding but a category mistake with consequences. www.rotecodefraktion.de/en/blog/llm-... #AI #LLM #Consciousness #AIEthics
No Ghost in the Machine: On mistaking language models for minds
ChatGPT sounds thoughtful. Claude sounds empathetic. And yet nobody is home. A mechanistic, philosophical and societal look at the difference between language models and consciousness.
rotecodefraktion.de
One call, and the model's gone. Three days after launch, the #Trump regime forced #Anthropic to kill #Fable 5 and #Mythos 5 — an export-control directive aimed at "foreign nationals." No court, no reason, no evidence. www.rotecodefraktion.de/en/blog/us-r...
Three Days at the Frontier: The Trump Regime Pulls Fable 5 and Mythos 5
Three days after launch, an export-control directive forces Anthropic to disable Fable 5 and Mythos 5 for all customers. What happened, why, and what it means.
rotecodefraktion.de
#n8n meets #SAP: Article 10 builds the bridge over OData and HTTP. A support pipeline enriches its tickets with master data from an SAP backend. The catch: SAP behaves differently from modern APIs — and when it fails, the key info hides where most give up. www.rotecodefraktion.de/en/blog/n8n-... #ai
n8n Meets SAP — the Bridge over OData and HTTP
The pipeline now enriches its tickets with master data from a SAP backend. This article shows the OData binding over HTTP, SAP's own error format, graceful degradation on an unknown customer, and wher...
rotecodefraktion.de
Local AI isn't automatically sovereign AI. apfel's coding agent runs on your Mac, no key. But the model is borrowed, on ground Apple can shift anytime. The agent isn't: built on the OpenAI protocol, the backend swaps out. www.rotecodefraktion.de/en/blog/apfe... #LocalAI #swift #ai
Sovereignty on borrowed ground
The local agent runs without a key and without the cloud. But what does that independence rest on? Two releases from the same week, a cloud handoff from Anthropic and Apple's MLX stack, bring the ques...
rotecodefraktion.de
Tools the agent doesn't write: MCP — Article 10 of the series "A Local Coding Agent with apfel". #apfel is the MCP client itself. What the agent builds instead: a small server of its own. #MCP #AppleSilicon #LocalLLM #Swift #AgenticCoding #AI www.rotecodefraktion.de/en/blog/apfe...
Tools the agent doesn't write: MCP
MCP moves writing tools out of the agent into separate servers. We build a small MCP server, attach it to apfel, and measure what MCP solves and what it doesn't. Article 10 of the series "A Local Codi...
rotecodefraktion.de
Two Models, One Entry Point, Round-Robin and Failover redundancy over Local AI Backends. Article 9 of the series "Getting Started with n8n". → www.rotecodefraktion.de/en/blog/n8n-... #n8n #AI #Failover #SAP #MLX #LLM
Two Models, One Entry Point — Round-Robin and Failover over Local AI Backends
A single AI backend degrades to the rule-based emergency exit on failure. This article builds real redundancy from two local models: round-robin in normal operation, automatic failover on a single out...
rotecodefraktion.de
The interactive terminal session, Article 9 of "A Local Coding Agent with apfel" From a one-shot command to a REPL. How a strict split of render logic and I/O makes a terminal interface testable, with no terminal at all www.rotecodefraktion.de/en/blog/apfe... #LocalAI #CodingAgent #Swift #apfel #llm
The interactive terminal session
The one-shot command becomes an interactive REPL session. The key to testability is the strict separation of render logic and terminal I/O. Article 9 of the series "A Local Coding Agent with apfel".
rotecodefraktion.de
#Anthropic makes the #Mythos class public, as Claude #Fable 5. A critical look at the benchmark records, the safeguard fallback to Opus 4.8, the June 23 pricing maneuver look like drug dealing, and the 30-day data retention. #ai #llm → www.rotecodefraktion.de/en/blog/clau...
Claude Fable 5: Anthropic's Mythos Class Goes Public, With Caveats
Two months after warning that Mythos was too dangerous to release, Anthropic makes the Mythos class public. A look at Claude Fable 5 and Mythos 5 between benchmark records, pricing tactics, and 30-day...
rotecodefraktion.de
#WWDC26 Platforms State of the Union at a glance. Foundation Models opens up to server models and goes open source, #CoreAI brings your own models on-device, and #Xcode 27 turns the coding agent into an actor that plans, builds, and tests. #ai #agenticcoding www.rotecodefraktion.de/en/blog/wwdc...
WWDC26 for Developers: Foundation Models, Core AI, and an Agentic Xcode
An overview of the WWDC26 Platforms State of the Union: the Foundation Models framework opens up to server models and goes open source, the new Core AI brings your own models on-device, and Xcode 27 p...
rotecodefraktion.de
Four Macs as a cluster, distributed inference and training with #MLX. How #RDMA over Thunderbolt, JACCL, and MLX run a trillion-parameter model locally across four M3 Ultras, and why that hardware is nearly impossible to get in 2026. → www.rotecodefraktion.de/en/blog/mlx-... #Apple #LocalAI #WWDC
Four Macs as a Cluster: Distributed Inference and Training with MLX
At WWDC26, Apple shows how MLX distributes large language models across multiple Macs. RDMA over Thunderbolt 5, the JACCL library, and a cluster of four M3 Ultras running a trillion-parameter model lo...
rotecodefraktion.de
Apple's on-device AI is fast & local but hits a ceiling on reasoning. Anthropic's new #Swift package hands off to Claude there @Generable typed output as the seam. Not #Claude in an app; the handoff becoming a pattern. #wwdc www.rotecodefraktion.de/en/blog/clau...
rotecodefraktion
AI, Apple Silicon, Coding-Experimente, SwiftUI und Haltung.
rotecodefraktion.de
One Entry Point, Two Classifiers, merging the AI Pipeline into the Live Path. Article 8 of the series "Getting Started with n8n". #n8n #AI #Ollama #llm www.rotecodefraktion.de/en/blog/n8n-...
One Entry Point, Two Classifiers — Merging the AI Pipeline into the Live Path
Until now two separate paths ran side by side: the rule-based ingest and the AI classification. This article merges them into one pipeline, makes classification a swappable stage, and positions the ru...
rotecodefraktion.de
Editing that works Article 7 of the series “A Local Coding Agent with apfel”. Why free tool calling fails at editing, and how constrained output takes the small model from 0/9 to reliable edits. → www.rotecodefraktion.de/en/blog/apfe... #LocalAI #CodingAgent #Swift #apfel
Editing that works: constrained output instead of tool guessing
The naive file edit fails on the small model. We build the agent around that weakness, with constrained output and a two-stage edit workflow, and measure how much can be recovered. Article 7 of the se...
rotecodefraktion.de
The local coding agent put to the eval — Article 6 of "A Local Coding Agent with apfel". 15 tasks, checked deterministically, no self-verdict from the model. Where it holds, where it breaks, and what it claims about itself. www.rotecodefraktion.de/en/blog/apfe... #LocalAI #Coding #swift #apfel #llm
The local coding agent put to the eval
A reproducible eval of 15 tasks across five categories measures where the local Foundation Model carries as a coding agent and where it breaks. With a cloud sample and a separate measurement of what t...
rotecodefraktion.de