Jason Lovell: LinkedIn Content Strategy & Writing Style

DeepSeek just showed us where the real AI bottleneck is. It's not where everyone thinks. They generated 1,800+ synthetic environments to train V3.2's agent capabilities. Not datasets. Environments: c…

There's a study from Stanford and Carnegie Mellon from last month that should be required reading for anyone deploying AI agents right now. The researchers put humans and AI agents head-to-head on re…

Someone just extracted what Anthropic calls the “soul document” from Claude 4.5 Opus. Not from a system prompt. From the *weights themselves.* This is fascinating for a reason few are discussing: we’…

Something quietly remarkable happened in computer use AI this year. And most coverage completely missed the architectural story. Eighteen months ago, the best models scored around 12% on OSWorld: the…

Mistral quietly shipped something today that deserves more attention than the headline specs. Yes, they released Mistral Large 3 (675B parameters, #2 on LMArena for open-source). Yes, it's multimodal…

For two years we tried to civilise LLMs with heavy extension machinery: plugins, MCP servers, complex client–server handshakes. Now Anthropic’s Claude Code Skills quietly suggest the opposite: drop a…