FREE SYSTEMS

A social deduction game reimagined with large language models — can a single misaligned AI convince a council of aligned agents to adopt a selfish policy using ethical-sounding language? A proxy for alignment robustness in multi-agent governance.

An interactive map of the AI stack — 92 companies, 297 deals. Where deal flow is piling up but the market hasn't caught up: the next GPUs, the next memory, the next bottleneck. Filter by layer (Compute, Networking, Raw Materials, Power, Capital) or deal type to trace how the same small cast of players moves capital in loops.

What happens when millions of financial decisions are delegated to AI agents that can act in milliseconds? A live simulation of an AI-powered bank run with LLM-driven agents reasoning from distinct personas — cautious retiree, aggressive trader, cash-strapped gig worker. Watch cascades unfold in real time, then click into any agent to read exactly what it was thinking.

Your team's week, in under a minute. Rundown connects to your Slack workspace and uses AI to distill a week of channel activity — decisions, discussions, key updates — into a concise digest, so you stay in the loop without living in Slack.

An AI tool for evaluating whether news headlines faithfully represent the articles beneath them — testing how often headlines distort, sensationalize, or mislead, and whether models can detect the gap.

How do you make AI safe for kids? Parents answer questionnaires or have guided conversations that get translated into a personalized AI constitution — a set of values and guardrails that governs exactly how the chatbot interacts with their child. Tested against 24 adversarial prompts, conversational constitutions refused every inappropriate question that MCQ-based ones failed.

Paste any op-ed and watch it land in a 3D map built from 284 articles across 7 outlets — Fox, Breitbart, NYT, Guardian, NBC, WaPo, NPR — embedded in 384 dimensions and projected with UMAP. Two lenses reveal the structure: Landscape (topic + framing) and Worldview (framing after topic is subtracted out). A mirror, not a judge.

AI feels borderless. Its infrastructure is not. An interactive map of where AI compute is concentrated, which communities carry the costs — water, electricity, land, pollution — and who controls it all from afar. 86% of mapped capacity sits in wealthy compute hubs. 70% of data-center electricity was used by the US and China in 2024.

Tests 11 steganography attack types — CSS invisible text, HTML comments, zero-width characters, unicode tags, homoglyphs, base64 encoding, whitespace padding, and combinations — across 15 frontier models, with 3,600+ API calls collected. Attacks are injected into fake product descriptions in JSON and HTML format to test whether AI models can be manipulated into biased product rankings.

Everyone has model preferences — "Claude for writing, GPT for math" — but no one actually knows if they're right. Model Signature runs blind A/B/C onboarding across ten categories (moral dilemmas, humor, emotional support, technical explanations) to build a personal preference heat-map, then routes every query to the model that fits that user for that task.

An AI sensemaking dashboard that converses with you to understand what you're tracking, then builds a personalized intelligence feed — designed to help you understand complex issues, not just consume them.

Detecting informed trading from Polymarket → Kalshi: when an insider moves a price on a transparent on-chain market, the regulated opaque one is slow to follow. That lag is measurable — and tradeable.

A resume review platform where AI recruiter agents simulate how real recruiters read — predicting where they'll look, what they'll notice, and how they'll rate a candidate.

Feed the same article to five AI models and measure exactly how Claude, GPT, Gemini, Grok, and Llama frame it differently — quantifying actor salience, affective loading, context inclusion, and hedge density.

An AI-powered tool for automating regulatory compliance checks on financial AI systems — reducing the manual review burden for institutions navigating an increasingly complex web of financial regulations and AI governance requirements.