doorss

Anthropic is donating Petri, its open-source alignment testing toolbox developed through its Fellows program, which can rapidly test any large language model for alignment properties.

The Anthropic Institute (TAI) outlines its research agenda, focusing on investigating AI's real-world impact from within a frontier lab and sharing findings publicly.

Anthropic presents research on natural language autoencoders that convert Claude's internal representations into human-readable text, offering a new window into how the model processes information.

A survey of 81,000 Claude users finds that workers in AI-exposed roles have greater concerns about job displacement, with concerns also higher among lower-income respondents.

Anthropic is fielding inbound investment offers that could push its valuation close to $1 trillion, driven by surging revenue from its Claude AI tool.

Despite recent improvements, AI sycophancy remains a persistent problem. A Stanford paper highlights how sycophantic AI advice continues to affect user decision-making.

Research examining how large language models are systematically distorting written language patterns, analyzing the measurable shifts in text produced with LLM assistance.

An examination of how AI-generated low-quality content is degrading the quality of discourse in online communities, overwhelming human-created contributions.

Simon Willison discusses how the boundary between casual "vibe coding" and professional agentic engineering is narrowing, raising concerns about quality and craft in software development.

Mozilla used access to the Claude Mythos preview to locate and fix hundreds of vulnerabilities in Firefox, providing detailed insights into the AI-assisted security hardening process.

Cloudflare is cutting over 1,100 jobs, roughly 20% of its workforce, as part of a restructuring amid the broader tech industry's shift toward AI-driven efficiency.

A critique arguing that AI coding agents like Claude Code optimize for code output rather than product quality, potentially leading teams to ship more code without better outcomes.

Google Chrome has been silently downloading a 4 GB AI model (Gemini Nano) onto users' devices without explicit consent, raising privacy and transparency concerns.

An investigation into how modern connected cars collect extensive driver data and feed it into advertising infrastructure, detailing the surveillance ecosystem built into vehicles.

Cambridge research finds that gambling advertisements on social media platforms disproportionately reach men, raising questions about targeted ad delivery and its societal impact.

Bluesky published technical details on how its "For You" feed is served, highlighting one of its most distinctive features: anyone can run their own custom feed implementation.

Bluesky experienced a sustained DDoS attack lasting nearly a full day, testing the resilience of the decentralized social network's infrastructure.

Britannia lisäsi pakotelistalle kaksi henkilöä, joiden epäillään yrittäneen horjuttaa Suomen vakautta välineellistetyn maahanmuuton avulla. Yle selvitti kaksikon taustat.

Suomesta on tullut satelliittien suurmaa, mutta Euroopalta puuttuu omaa laukaisukykyä. Ruotsiin rakennetaan nyt kipeästi kaivattua laukaisuasemaa.

Uber Eats tulee Suomen markkinoille ja haastaa Woltin valta-aseman. Lähetit saavat valita työsuhteen tai yrittäjämallin.