ChippyTech Blog

I didn’t expect to stumble into anything particularly alarming. Like a lot of people experimenting with local AI, I spend time poking around tools, testing models, and occasionally checking how widely they’re being adopted. Out of curiosity more than anything, I ran a few queries through Shodan—the search engine that indexes internet-connected devices. What came back wasn’t just interesting. It was unsettling. According to Shodan, there are currently: 34,953 publicly exposed Ollama instances 1,308 publicly exposed llama.cpp instances Not just visible— unsecured . What This Actually Means Both Ollama and llama.cpp are designed primarily for local inference . They’re fantastic tools: lightweight, fast, and increasingly powerful. The assumption is simple—you run them on your own machine or internal network. But many of these instances are sitting directly on the public internet with: No authentication No rate limiting No access controls Open APIs ready to accept requests from anyone In ot...

The web is ephemeral by design. Pages vanish, domains expire, content shifts or gets rewritten. Yet for researchers, journalists, developers, and everyday users, the ability to access historical versions of the web is essential. But one question has remained surprisingly underexplored: which web archives are actually being used at scale? To answer that, we turned to one of the largest structured knowledge repositories on the internet—Wikipedia—and conducted a large-scale analysis of how it references archived content. How We Built the Dataset To get a meaningful picture of archive usage, we needed a dataset that was: Massive Diverse in subject matter Curated and moderated Rich in external references Wikipedia checks all those boxes. We analyzed 187,817,169 external links extracted from a full Wikipedia dump. This includes links across articles, references, citations, and templates—essentially every outbound URL that Wikipedia points to. From there, we: Parsed all ...

Tasker has always been one of the most powerful automation apps on Android. It lets you create workflows where your phone reacts automatically to events — like turning on Wi-Fi when you get home, or sending a message when you miss a call. But now, with the GenAI Tasker plugin , you can take things much further by adding artificial intelligence into your automations. This means your phone doesn’t just follow rules anymore — it can think, respond, and generate content dynamically . What Is the GenAI Tasker Plugin? The GenAI Tasker plugin is an add-on for Tasker that lets you connect to multiple AI providers and use them directly inside your automation flows. It supports a wide range of AI platforms, including: OpenAI Anthropic Google OpenRouter Ollama Grok This flexibility means you can choose the model, pricing, and performance that best fits your needs — or even switch between them depending on the task. Why Add AI to Tasker? Adding AI transforms Tasker f...

In the summer of 2025, OpenAI did something that many thought would never happen again: they opened the gates. After years of pivoting toward increasingly closed proprietary systems, the release of gpt-oss —specifically the 120b and 20b models—marked a tectonic shift in the AI landscape. While the tech world often obsesses over the "next big thing" in RLHF (Reinforcement Learning from Human Feedback), gpt-oss is a reminder that sometimes, the best gift isn't a perfectly polished, hyper-aligned mirror of human preference. Instead, it’s a raw, high-reasoning engine that gives the power back to the developers. The Power of the "Under-Aligned" The most striking feature of gpt-oss isn't its benchmark scores (which, for the record, are stellar), but its limited RLHF . For the uninitiated, RLHF is the process that makes an AI sound "nice," safe, and helpful. However, heavy RLHF can also lead to "model collapse," where the model becomes overly...