Thinking Machines just open-sourced an AI model — and that changes what skills your family actually needs

A Philippine AI company called Thinking Machines released an open-weights language model named Inkling this week — small enough to run locally, openly licensed, and built to work well in Southeast Asian languages. The Hacker News thread lit up immediately. Most of the conversation was technical. Almost none of it was about what families should actually do with this information.

So let's do that.

What's actually changing

The Inkling release is not, by itself, a household event. One more model entering an already crowded field doesn't reorder your week.

The pattern behind it does.

Open-weights models — AI systems whose underlying parameters are publicly released, so anyone can run them without paying a subscription or pinging a corporate server — have gone from rare to routine in roughly 18 months. Each new release tends to be smaller, cheaper to run, and more capable than the last. Inkling is notable partly because it comes from outside the U.S.-European axis that has dominated this space, which suggests the diffusion is genuinely global now, not a Silicon Valley phenomenon trickling outward.

What this means for a household: within a reasonable planning horizon, a family with a mid-range laptop will be able to run a useful AI assistant with no internet connection, no subscription, and no data leaving the house. That is a genuinely different situation than the cloud-dependent tools most people are using today.

It also means the skill premium that AI currently provides — knowing which tool to use, how to prompt it, how to verify its output — is compressing fast. The advantage goes to people who can pair AI fluency with judgment and domain knowledge, not people who merely have access to the tools.

What we'd actually do

Audit which skills in your household you've been offloading to AI subscriptions, and which ones you've let atrophy because you assumed AI would always be there.

This sounds abstract until you make a list. Can someone in your home read a contract without pasting it into ChatGPT? Navigate a map without a route-planner? Write a coherent complaint letter? The question isn't whether AI assistance is bad — it isn't — it's whether you've quietly traded a skill for a dependency without noticing. Dependencies are fine when the service is reliable. They're problems during outages, account suspensions, or price spikes.

Learn to run at least one local AI tool before you need it.

Ollama is a free, open-source application that lets you download and run models like Llama or Mistral on a home computer without internet access. The setup takes under an hour for anyone comfortable installing software. The point isn't to abandon cloud tools. The point is to have a fallback that works when your internet is down, when a subscription lapses, or when a service shuts its free tier. Recent BLS consumer expenditure data shows software subscription costs are a growing share of household budgets — having local alternatives matters.

Prioritize teaching children verification skills over access skills.

Every kid with a smartphone already has access to AI. What most of them are not learning is how to check whether an AI answer is true. Lateral reading — opening multiple tabs, checking primary sources, finding the institution or person behind a claim — is a teachable skill that compounds. A child who can verify AI output will outperform a child who merely uses AI fluently, especially as the tools become commoditized.

Run a one-week household experiment: log every time someone in your home uses an AI tool, and note whether it replaced a skill or extended a skill.

Replacing: using AI to write your kid's thank-you notes. Extending: using AI to draft a letter you then edit, personalize, and send. The log will probably surprise you. It tends to reveal a few high-value use cases and a larger number of small dependencies that felt convenient but weren't building anything.

The bigger picture

The proliferation of open-weights models is good news for household resilience. Tools that run locally, cost nothing, and don't require a corporate intermediary are, in preparedness terms, more robust than cloud-dependent alternatives. Thinking Machines releasing Inkling is a data point in a trend toward decentralization that, if it continues, gives households more options, not fewer.

The risk isn't the tools themselves. The risk is the assumption that access to AI is equivalent to capability — and that building real skills is therefore optional. It isn't. Durability means you can function when the tools work, when they're slow, and when they're gone. The tools are getting better and cheaper. That's no reason to stop sharpening the person using them.