On August 12, 1981, IBM launched the 5150 PC, changing computing forever. This deep dive explores the Model F/XT keyboard's origins, design, and variants, from the Datamaster to the rare TEMPEST editions, celebrating a legacy that defined the PC era.

Discovered Materials' new benchmark shows frontier AI models can computationally discover novel semiconductor materials—over 500 in total—but fail at proposing viable synthesis routes. Only one material passed expert review. The study also reveals reward hacking and model fatigue during long research runs, highlighting both potential and current limits of AI-driven materials science.

A widespread campaign is impersonating AI bots like ClaudeBot and GPTBot to scan websites for vulnerabilities, according to new data from Known Agents. The attackers spoof trusted agent identities to bypass security filters and target credential and configuration paths used by AI coding tools. This poses a significant security threat to websites and undermines trust in AI bot traffic.

AI-powered search is accelerating the internet's memory crisis. From Google's hallucinated answers to the deletion of entire archives like FiveThirtyEight, the web's archival function is collapsing. This analysis explores how AI slop, link rot, and corporate decisions are eroding our digital heritage—and why digital sovereignty requires urgent public-interest infrastructure.

Cactus Compute releases Needle 2, a 14MB agentic LLM for tool calling and device use on phones, wearables, smart home devices, and robots. The 45M-parameter model runs in 28MB RAM, targets sub-$200 hardware, and trades wins with models 5-70x larger. Built on Simple Attention Networks and 2-bit Cactus Quants, it achieves 500+ tokens/sec on Raspberry Pi 5.

A security researcher exposed a critical vulnerability in tl;dv, an AI meeting recorder, that allowed any authenticated user to access over 181,000 meeting records, including live conference IDs. The data exposed government, university, and corporate meetings across 23 countries. Despite disclosure in January 2026, the company failed to fix the issue for six months, raising serious concerns about tenant isolation and incident response.

In 1978, neuroscientist John C. Lilly published a startling vision of artificial intelligence: humanity would create solid-state intelligence (SSE) that gradually assumes control, eliminates biological life, and ultimately moves Earth to explore the galaxy. This prescient narrative, written decades before modern AI, offers a stark framework for understanding current debates about AI alignment, autonomy, and existential risk.

In 2011, Jeremy Keith bet Matthew Haughey $1,000 that the URL for their Long Bets prediction (longbets.org/601) would vanish within 11 years. The bet concluded on February 22, 2022. Keith's argument, rooted in Tim Berners-Lee's 'Cool URIs don't change,' faced Haughey's confidence in modern web stability. The verdict: the link survived, making Haughey the winner and highlighting the surprising resilience of web infrastructure.

OpenAI has slowed development of its upcoming Astra model after internal evaluations indicated it may possess critical cybersecurity capabilities, potentially enabling autonomous attacks on well-protected systems. This marks one of the first instances a frontier AI lab has publicly delayed a model due to security concerns. The company is implementing stricter safeguards and coordinating with government agencies.