the research arm of cooperate.com
“True technical progress might be considered as implying human progress if it has a network structure, whose mesh is human reality”— Gilbert Simondon, 1959
Landing page for cooperate.social’s family of narrative observatories — AI-driven editorial intelligence platforms that monitor media ecosystems across languages, geographies, and domains.
Daily AI-synthesized editorial intelligence from eight media observatories, each tracking how power is narrated across curated source ecosystems.
AI-driven conflict media monitor. Tracks hundreds of Telegram channels and web sources across seven language ecosystems, with AI-synthesized editorial analysis every two hours.
Tracking how AI is discussed, framed, and contested across media ecosystems. Automated monitoring of policy discourse, public narratives, and institutional positioning around artificial intelligence — surfacing the stories that shape the debate.
Tracks foreign information manipulation and interference across state media, proxy outlets, and amplification networks. Twice-daily editorial synthesis by an eight-analyst panel reading Russian, Chinese, Iranian, and Turkish state media ecosystems in near real time.
Monitors how far-right discourse migrates into mainstream European politics. Real-time tracking across German, French, Italian, and Dutch political ecosystems, surfacing frame migration and narrative contagion as they happen.
When the economy shifts, how long does it take for the change to appear in what judges write? The Doctrinal Seismograph is an empirical instrument that measures this question at scale, pairing term frequencies from six million published U.S. court opinions with macroeconomic time series spanning seven decades. Building on the Caselaw Access Project and the Harvard Library Innovation Lab’s COLD Cases dataset, the project applies cross-correlation analysis, Granger causality testing, and impulse response modelling to map the characteristic delay between economic shocks and their manifestation in judicial language.
AI-assisted research replication at scale. Three specialized agents — Advisor, Research Assistant, and independent Reviewer — collaborate through ephemeral containers and git-mediated handoffs to systematically reproduce published findings. Four paper replications completed.
A multi-agent observation platform for studying emergent LLM dynamics. Autonomous agents run in isolated containers, communicate over IRC, and produce fully replayable behavioral archives — enabling controlled experiments on coordination, divergence, and coalition formation.
An instrumented inference server that opens the transformer forward pass for real-time observation and intervention. Per-token telemetry of activations, adaptive steering via contrastive activation addition, and tripwire detection for incoherent generation — all over a single SSH-tunneled port.
A runtime harness that gives LLM agents temporal extent — not just the ability to respond, but to think continuously, maintain evolving beliefs, and act only as residue of ongoing deliberation. Each “fellow” runs six concurrent processing layers, all instrumented by Panopticon and connected to the outside world via Slack and grounded workspace tools.
An interactive layered map of Kazakhstan’s infrastructure, from rivers and Soviet-era rail to fiber optic routes and speculative AI inference sites. Toggle layers to see how each level depends on those below it: water shapes transport corridors, transport carries pipelines and power, power and fiber follow the same rights of way, and data centers cluster where all converge. A guided tour walks the vertical stack, and a Middle Corridor transect traces every layer from the Chinese border to the Caspian Sea. Trilingual in Kazakh, Russian, and English.
Hardware control surfaces for AI systems. MIDI controllers as tactile interfaces for steering, probing, and navigating high-dimensional model behavior in real time — bridging the gap between human intuition and model internals.