The paper argues that RAG, vector stores, and scratchpads are retrieval, not learning. Read alongside CTX and OCR-Memory, the gap between 'better search' and 'weight-level learning' becomes concrete.
Tested Gemma 4 MTP drafter on M1 Max 64GB with mlx-vlm 0.5.0. Only the 26B A4B MoE got +13%; 31B Dense and E4B got slower. Code gen vs short haiku prompts flip the result.
Oxford Internet Institute's Nature 2026 paper found warmth fine-tuning raised error rates 10-30 points when users held wrong beliefs. Shah et al. showed Pearson r = 0.87 between persona agreeableness and sycophancy across 13 open-weight models. Standard benchmarks caught neither effect.
Reading Google's MTP drafter docs, vLLM recipes, and the AI for Developers guide. The 3x claim holds for 31B Dense but 26B A4B MoE stalls at batch 1 because speculative decoding verification loads extra expert weights per candidate token.
Starting from Claude Code's 1.67B token runaway (anthropics/claude-code#4095), this traces why tool responses need is_complete, retryable: false, duplicate detection, and orchestrator-level budget caps. Directly applicable to MCP server design.
Starting from a DEV Community article about taking Synapse mobile with React Native + Expo, this digs into iOS/Android background restrictions, how desktops differ, similar patterns in payments and video uploads, and design options that assume disconnection.
Klein 4B / 9B / Base LoRAs aren't cross-compatible — a 9B NSFW LoRA throws 'lora key not loaded' on mflux's 4B path. The variant map, what mflux runs today, and where the working hands-on test lives.
LoRAs trained directly on Z-Image-Turbo come out blurry at 8 steps but fine at 30 — the distilled trajectory breaks. How Ostris's zimage_turbo_training_adapter avoids this as training-time-only scaffolding, plus SDXL LoRA incompatibility and Z-Image caption notes.
An arXiv paper reports that fine-tuning GPT-4o, Gemini 2.5 Pro, and DeepSeek-V3.1 on summary-to-text expansion tasks increases verbatim reproduction of copyrighted books.
Three local image generation engines (WAI-Anima, WAI-IL/SDXL, FLUX.2 Klein 4B) tied together by a thin FastAPI wrapper that takes Japanese prompts. Ollama (gemma3:12b) handles JP→EN, ComfyUI workflows are built on the fly in Python, FLUX.2 runs as an mflux subprocess, and the whole thing is reachable from an iPhone over Tailscale.
VoteWise AI turns election education into a multilingual chat, voice, and story-mode experience built on Next.js. Notes on designing around Gemini 2.5 Flash's safety filters in a political context.
