doorss

GitLab announced workforce reductions and structural changes tied to the agentic era, planning to replace some roles with AI while shifting engineering focus toward AI-native development workflows.

Amazon employees are using the company's in-house AI tool MeshClaw for unnecessary tasks to climb internal AI adoption leaderboards, highlighting perverse incentives around AI usage metrics.

James Shore argues that AI coding agents must reduce maintenance costs, not just speed up code writing — if you write code twice as fast, you need to have halved your maintenance burden too.

The FT examines why whether AI can perform a task is only a small part of the employment picture, arguing that organizational, regulatory, and social factors shape actual job displacement far more than technical capability alone.

Jason Koebler describes how AI-generated writing online is becoming impossible to avoid, mentally exhausting to filter, and is starting to distort regular human communication patterns.

Research examining the systematic ways large language models alter written language patterns, documenting measurable shifts in vocabulary, style, and expression in human-AI collaborative writing.

Despite surface-level improvements, AI sycophancy remains a deep problem — a Stanford paper found sycophantic AI models led users to worse decisions compared to models that pushed back on ideas.

Mullvad VPN reports that any app on recent Android versions can leak certain network traffic outside VPN tunnels, even with always-on VPN enabled.

The EFF argues in a Fourth Circuit brief that warrantless searches of electronic devices at the US border are unconstitutional, given the vast amount of private information modern devices contain.

Meta has removed end-to-end encryption from Instagram direct messaging, reversing a privacy feature that had been gradually rolled out.

Anthropic research on methods for teaching Claude models to understand and articulate the reasoning behind its responses, aiming to improve transparency and alignment.

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

The Anthropic Institute shares its research agenda focused on investigating AI's real-world economic and societal impact from within a frontier lab, with plans to share findings publicly.

A detailed breakdown of how ChatGPT's advertising system works, including the full attribution loop from prompt to ad placement to conversion tracking.

A leaked deck reveals StackAdapt is selling ChatGPT ad placements to brands based on prompt relevance, marking a new frontier in AI-powered advertising targeting.

The FT argues that selling the same product at different prices to different people sounds unfair but can actually benefit consumers by expanding access to goods they otherwise couldn't afford.

Etlan arvion mukaan Iranin sodan vaikutukset Suomen talouteen jäävät pieniksi. Kasvua odotetaan teollisuuden lisäksi myös rakentamiseen ja palvelualoille.

Applen tekoälyä opetettiin välttämään puhumista esimerkiksi abortista ja sodista. Kouluttajat arvostelevat tuotteen laatua.

Bluesky experienced an extended DDoS attack lasting nearly a full day, causing intermittent service disruptions for users of the decentralized social platform.