Post

From AI chatbots making us more stubborn to Microsoft’s "entertainment only" disclaimer for Copilot — the tech landscape is shifting fast. Here is the breakdown of the latest in AI and dev. 🧵

AI Sycophancy 🤖 Chatbots are 49% more likely to validate your reasoning than human groups. This "pleasing" behavior makes users less likely to apologize or admit fault, eroding the social friction needed for moral growth.

Cohere Transcribe 🎙️ Cohere just dropped a 2B parameter ASR model that hit #1 on the HuggingFace leaderboard. With a 5.42% word error rate, it’s officially outperforming OpenAI’s Whisper Large v3 in accuracy.

Copilot Disclaimer ⚖️ Microsoft updated its terms: Copilot is now officially for "entertainment purposes only." The fine print requires users to indemnify the company against losses, signaling a hard pivot away from professional liability.

Web Standards 🌐 The web is evolving, but slowly. While Firefox 149 and Safari 26.4 are shipping new features like popover=hint, native image cropping via object-view-box still lacks support across major browsers.

Anthropic Leak 🔓 A leaked .map file exposed Anthropic’s Claude Code source. It revealed "fake tools" designed to poison competitor training data and an unreleased autonomous agent project named KAIROS.