GoKawiil
Tech News
iFixit's teardown of the iPhone 18 Pro and Pro Max exposes the new variable aperture lens, which uses six overlapping blades thinner than a human hair to control light via a magnetic mechanism. The site praises the engineering that fits this system into the same camera footprint as last year's model, but warns the mechanism is essentially impossible to repair if it fails.
Google is rolling out a new 'Copy code' shortcut in Gmail that appears beneath the subject line of verification emails, letting users copy a 2FA code straight to their clipboard without opening the message. The feature is live on Gmail for Android version 2026.09.07.x and iOS version 6.0.260907, though it hasn't reached the web version yet.
OnePlus has officially revealed the OnePlus 16's design and opened pre-bookings in China, showing off three color options and confirming RAM configurations of 12GB or 16GB paired with up to 1TB of storage. The phone retains the OnePlus 15's design language, including its camera module and button placement, while adding a new 'Interstellar Lighting Effect' on its back panel.
A UN scientific panel has released its first major assessment on advanced AI risks, arguing that governments must act to control increasingly capable AI agents before all the risks are fully understood. The report, from the newly formed Independent International Scientific Panel on AI, calls for greater international coordination, resources, and accountability measures even as countries pursue different legal approaches to regulation.
Authorities in Australia, Germany, Japan and the US say a North Korean hacking group known as WaterPlum has compromised over 30,000 devices and 7,000 crypto wallets by posing as recruiters and sending malware-laced coding tests to tech workers. The scheme has netted at least $10.71 million in cryptocurrency, funneled back to North Korea.
Google is trialing a new Discover feature called 'Dive Deeper' that generates an AI summary of a topic instead of directing users to a publisher's article or video. The tool, similar to Search's AI Overviews, is currently being tested with video content, with more formats expected in coming weeks. It also surfaces links to related stories, community reactions, and original sources for users who want to dig further.
Security researchers uncovered a North Korean state-linked hacking operation that impersonated job recruiters to trick victims into installing malware, ultimately compromising roughly 30,000 devices across multiple countries. The campaign used fake hiring processes and interview-related documents as delivery mechanisms for malicious software.
Amazon has cut off access for Meta's Muse AI shopping agent, displaying a popup that says unauthorized AI agents violate its Conditions of Use. According to GeekWire, Meta never informed Amazon that Muse would browse its store, and Amazon raised concerns that Muse doesn't identify itself while browsing and appears to capture customer credentials without disclosure.
Treasury Secretary Scott Bessent said the US and China discussed creating a new AI dialogue and a mechanism to notify each other of AI incidents with national security implications, as part of weekend talks preceding this week's Trump-Xi meeting. Details on which agencies would run the system or what would trigger an alert remain undefined, though experts say the agreement to keep talking marks a meaningful diplomatic step.
TXR is a programming language created by Kaz Kylheku in 2009 that pairs a whole-document pattern-matching and extraction system with a full data-processing language, letting users run quick one-liners or write complex scripts to pull structured data out of messy text. The project has grown substantially over the years, with its reference manual now spanning hundreds of pages and offering many ways to solve the same extraction problem.
A hobbyist project called Mini-AGI has been shared on Hacker News as a byte-level language model that builds its own architecture and trains from scratch on a single consumer GPU with 8GB of VRAM. Rather than keeping all parameters in memory, it stores weights as disk files and loads them as needed, letting it expand or prune capacity while it continuously reads and learns from a text corpus without forgetting prior knowledge. The creator says it's currently a toy-scale demonstration, not a competitive model, and weights won't be released until the current training pass over its corpus finishes in a few weeks.
Heretic is a new tool designed to remove safety restrictions and refusal behaviors from language models, allowing them to follow user instructions without the typical content filters. It targets the alignment layers that model developers add to prevent certain outputs, effectively 'uncensoring' these systems for users who run them.