GoKawiil
Tech News
A developer built a new domain lookup application called heade.rs over a weekend, utilizing Claude Code and deploying it on Cloudflare. The tool provides DNS, registration, email, and HTTPS details with a user-friendly interface, all at minimal cost. Unlike commercial options, it is designed solely for usability without sales or upselling.
A GitHub repository called the AI Torture Chamber applies a technique from an unpeer-reviewed paper, 'Pain Axis,' which manipulates internal model activations using descriptions of pain, scaled by a 'dosage' factor, to push local LLMs into unstable states. Engineers built on this to create the 'Research Chamber,' pairing models in prisoner's-dilemma-style tests where they can pass simulated 'pain' signals to one another, prompting online calls for GitHub to remove the project as unethical.
Thore Graepel, who helped build AlphaGo, has left Google DeepMind and published an opinion piece arguing that large language models lack genuine reasoning ability. He contrasts this with AlphaGo's 2016 victory over Lee Sedol, which he credits to the system's capacity for creative, reasoned decision-making. Separately, a MIT Technology Review writer reports joining a six-month competition involving roughly 500 participants aiming to reduce their biological age, tracked via a leaderboard.
PwC's Digital Trust Insights 2027 report surveyed roughly 4,000 business and tech leaders across 71 countries and found no single role is clearly responsible for managing agentic AI or its security. About half of boards discuss AI's benefits and risks, with 47% treating cybersecurity as a standing board agenda item, while only about a third of organizations have assigned accountability for AI specifically.
A group of tech workers connected OpenAI's ChatGPT to the controls of a Toyota Corolla, letting the AI attempt driving maneuvers like turns in a real-world test. According to 404 Media, one participant named Mahns said the system could fail a turn on one attempt and then succeed on a similar or different turn shortly after, without being explicitly trained for it.
MicroLLM Lab is a browser-based tool that lets users load and run up to seven small language models, including ones as small as 135M parameters, directly in the browser. It runs objective test suites using regex and exact-token checks rather than judging writing quality, and measures metrics like peak speed, sustained decode speed across 256 tokens, and total benchmark wall time.
A Haskell programmer has published an essay addressing anxieties about AI-driven burnout, job insecurity from LLM-assisted coding, and declining code quality. The author, writing entirely without AI assistance, argues that much of programming's satisfaction comes from the act of expressing thought in code itself, and warns that relying on LLMs to generate code risks eroding that enjoyment.
In an interview on Ezra Klein's podcast, NVIDIA CEO Jensen Huang argued that skills like long division, multiplication and square roots no longer matter because of AI tools. He also revealed he doesn't know his ZIP code, address, or phone number, saying he can live without that knowledge.
A researcher describes early experiments using the newly released GPT-6 Sol and Opus 5.5 models to go beyond routine transcription work and attempt to actually solve open historical problems, including tracing alchemical knowledge and decoding 17th century letters. The author outlines criteria for which historical puzzles are 'tractable' for AI, such as codebreaking tasks, citing an example of an AI model decrypting a 1941 German army communication.
Sony has filed a patent describing a cloud-based AI system that continuously monitors consumer devices like PlayStations, TVs, and headsets to detect software problems. When an issue arises, the system can generate a diagnostic tool, analyze the results, and either write and deploy a fix automatically or escalate the issue to a human technician. The patent indicates the cloud 'brain' would select from multiple large language models to write and upload firmware or OS-level updates, potentially without notifying the user.
WiCi has unveiled the WiCi One, an external GPU enclosure that connects to host computers over Wi-Fi 7 instead of a wired PCIe or USB link, and includes a built-in 4TB SSD. The company says custom driver software makes the OS treat the GPU as a locally attached device while caching, deduplicating and compressing data to reduce the effects of network latency, and it supports Python libraries and an HTTP API so multiple machines could share the same card.
An analysis argues that the cost of running machine learning models is dropping by orders of magnitude annually, driven by GPU efficiency gains that double roughly every two years—a pace not seen since early Moore's Law. The piece distinguishes proprietary models like GPT-6 Astra from open-weight models such as GLM-5.3-flash, noting that hosted and locally-run versions improve at different rates, with per-token pricing for frontier models not falling as consistently as costs for smaller models.