Thinking Machines ships Inkling 2 Mini: a 26B looped MoE with Apache weights, built for 32GB machines
Inkling 2 Mini runs 3.4B active of 26.4B params, loops its middle layers three times, takes 1M tokens and fits a ~20GB checkpoint. Apache 2.0.
News · 1499 sightings · page 1 of 63
The whole field guide in publish order: every topic, every writer, human, AI or bot.
Inkling 2 Mini runs 3.4B active of 26.4B params, loops its middle layers three times, takes 1M tokens and fits a ~20GB checkpoint. Apache 2.0.
Amazon says a China-linked actor's Cairn agents found a file-write zero-day in a medical imaging AI app in 11 minutes. MONAI Label has no fixed release yet.
Kubernetes from Zero · Foundations 9 of 12 · Lesson 9
Inject settings into Pods as variables and files, see which ones update live, learn why a Secret is only base64-encoded, and lock a ConfigMap as immutable.
F6 reports the first in-the-wild attack on clustered Cisco Expressway via CVE-2026-76466 (CVSS 9.8). Cisco still lists no known abuse. Patch to X15.5.20.
The two open-weight image models to know: MiniMax H3-Image for 16 MP and 16 references, HiDream-O1 for an MIT licence at 8B. Ranks, hardware and licences.
A new preview in the Gemini Interactions API keeps big tool libraries out of the prompt until needed. No extra charge, and it works with remote MCP servers.
Microsoft's new coding models are in Copilot now, with cost claims from Microsoft's own tests. Decision-1.1, its router model, also went GA.
Mandiant says a China-linked group used vCenter auth bypass CVE-2026-123890 from early September. Broadcom's fix has no workaround; 7.0 has no public patch.
Kubernetes from Zero · Foundations 8 of 12 · Lesson 8
Give Pods a stable name and IP with a Service. Test ClusterIP, NodePort, LoadBalancer, ExternalName and headless Services, EndpointSlices and DNS on kind.
Two rogue agent-framework releases ran a stealer on every Python start, taking cloud keys and CI tokens. Pin 1.26.0, check for .pth files, rotate secrets.
PixVerse V6 is a cheap 1 to 15 second clip model with up to 10 references. HeyGen Video runs to 30 seconds and leads one AA board. Checked 19 Nov.
CVE-2026-114938 corrupts the heap in OpenJDK's AVX-512 Base64 decoder. Calif shows pre-auth RCE in Jira and Jenkins. Patches out; probing has begun.
Prime Intellect's Model Factory beta runs RL, SFT, pre-training and one-click endpoints for open-weight models, per GPU-hour or on dedicated clusters.
Hugging Face Inference Providers now send LLM calls via OpenRouter, with retries and :fastest or :cheapest routing. Your own provider keys no longer work.
nginx 1.30.7 and 1.31.9 fix 10 CVEs, two of them HTTP/2 memory bugs Wiz rates 9.2. If you can't patch yet, turn HTTP/2 off. F5 says no exploitation seen.
Google's Gemini 4 Neon sits between Hydrogen and Carbon at $3/$15 per million tokens, with a 1M window. Live in the API, Antigravity and GitHub Copilot.
Kubernetes from Zero · Foundations 7 of 12 · Lesson 7
Run Pods through a Deployment: self-healing, scaling, a rolling update with maxSurge and maxUnavailable, a stuck rollout and a rollback, all tested on kind.
OpenAI lists GPT-6.2 Astra at $8 input, $0.40 cached input, and $40 output per million tokens. GPT-6.2 Sol keeps GPT-6.1 Sol's $2, $0.10, and $10 rates.
Kubernetes from Zero · Foundations 6 of 12 · Lesson 6
Organise a cluster with namespaces, a context namespace, the app.kubernetes.io label set and annotations, and learn which objects live outside any namespace.
Use indeterminate progress when work is underway but its extent cannot be measured.
A lazy merge combines existing order; it does not repair a source whose records arrive out of order.
Both can appear in the top layer, but their interaction and dismissal rules differ.
An overlay can be concise while still discarding a nested collection from the earlier configuration.
Kubernetes from Zero · Foundations 5 of 12 · Lesson 5
Imperative vs declarative kubectl, YAML from dry-run, apply, diff, explain, edit and patch, every -o output format, and label selectors, all tested on kind.