iFixit Launches a Mobile App for On-the-Go Repair Work
Bundles an LLM-powered chatbot, FixBot, which Kyle Wiens admits "sometimes gets things wrong."
Right-to-repair campaign group, repair guide platform, and toolmaker iFixit has announced the launch of a dedicated mobile app β capable of monitoring your phone's health, pulling up guides, and, of course, featuring a large language model (LLM) chatbot.
"Today, you can download the iFixit app from the App Store and Google Play Store," iFixit's Kyle Wiens says, a move that follows the company's 2011 iPhone app launch and its subsequent removal from the App Store in 2015 in, Wiens claims, retaliation for the publication of an Apple TV developer unit teardown. "All the repair guides you know and love, streamlined for mobile. A workbench that keeps track of your repairs. And some awesome app-only features, including a battery lifespan predictor and voice control with our brand-new AI repair buddy, FixBot."
The app's primary functionality is, of course, in providing access to teardown and repair guides for thousands of devices, some created by iFixit itself and others contributed by community members. Secondary to this is a tool designed to monitor the battery health of the phone on which it's installed. "The app tells you months in advance when your battery will need replacing," Wiens claims, "so you can plan the repair on your schedule instead of scrambling when your phone dies at the airport."
A more controversial inclusion is "FixBot," a large language model-powered chatbot. "You tell it whatβs happening: your phone dies at 30%, your washing machine won't drain, your mower sputters and stalls," Wiens says of the new tool. "It asks follow-up questions. It eliminates possibilities. It thinks out loud with you, the way a master technician would, until the diagnosis clicks into place. Then it finds the parts and walks you through step by step."
Of course, "thinks" and "asks" and even "finds" are terms arguably misused here: large language models operate by turning user prompts into "tokens," then statistically selecting the most likely continuation tokens from its training data β forming something that looks like an answer, but which has not actual thought or reasoning behind it. The result is that LLMs can, and do, "hallucinate" β providing answer-shaped objects in full confidence which are entirely divorced from reality.
When you're asking the chatbot how to repair something safety-critical, like the brakes on your car, or which can expose you to high voltages, like a TV set, that's a major problem β and one which, Wiens admits, is entirely unsolved. "FixBot is an AI, and AI sometimes gets things wrong," Wiens says in distinct understatement. "We've worked hard to keep it grounded in facts; every answer pulls from repair guides, teardowns, and real service manuals rather than internet hearsay. But it's not perfect, and we'd rather you know that upfront than find out mid-repair."
Despite this, and despite ethical concerns regarding the massive environmental impact of training and inference plus the unethical nature of capturing copyright and not-for-commercial-use material to be minced up and reconstituted as a slurry of tokens, iFixit has made the chatbot available to the public in its new app β free for now, then with a paid tier planned for the future.
The iFixit app is now available on the Google Play Store and Apple App Store.
Freelance journalist, technical author, hacker, tinkerer, erstwhile sysadmin. For hire: freelance@halfacree.co.uk.