Ollama 0.5 Ships Multimodal LLaVA Support
Pull `ollama run llava` and the new image-understanding pipeline is up — vision encoder swap, structured-output mode, and a fast CLI for batch captioning.
The curated front page — featured + latest across every section.
Off Screen Space covers AI and the slower technologies it touches — home, aging, work — with editorial care and a wiki-deep archive.
Releases and announcements over the next 90 days.
Pull `ollama run llava` and the new image-understanding pipeline is up — vision encoder swap, structured-output mode, and a fast CLI for batch captioning.
The desktop UI's server mode now mirrors OpenAI's `/v1/chat/completions` shape end-to-end — drop-in `OPENAI_BASE_URL` swap from the Python SDK, no auth shim required.
Mistral AI publishes the 123B-parameter weights under Apache 2.0 — Codestral-class reasoning at half the GPU footprint of Llama 3.1 405B. Locally runnable via vLLM, llama.cpp, and MLX.
We ran the same 4-bit quant on both backends across coding, summarisation, and long-context recall. MLX wins single-prompt latency; llama.cpp wins throughput. Full numbers + memory traces inside.
From an Ollama frontend to a polished multi-user RAG platform in 14 months. We talked to the maintainers about scaling moderation, the plugin marketplace, and what's next.
Hands-free AI browsing arrives on iPhone and iPad. Point your camera, ask Copilot questions, and get spoken answers. We walked the feature through five real use cases.
Anthropic's new flagship model adds a 'thinking' mode that visibly reasons through problems before answering. We tested it on long-form coding, planning, and research tasks.
A practical guide to using Projects to give Claude persistent context about your work. The setup that finally made AI feel like an assistant who remembers you.
The Connectivity Standards Alliance finalized Matter 1.4. Here's what changes — including the new AI routine layer — and what works on day one.
After years of false alarms with smart-home gadgets, the simplest setup was the one that actually changed my father's day.
Benchmarks show R2 matching GPT-4o on coding while using 40% less VRAM. Local deployment is now viable on consumer hardware for the first time.
Releases and announcements over the next 90 days.
Tap to follow — manuals, photos, and walk-throughs from these devices show up in your feed.