Silicon Society
Updated 2026-10-20
Method

How this works

This is a tracker, not a news outlet: it doesn't write stories, it finds and organises public ones. Here's exactly what it reads, how it decides what's relevant, and what it can't see.

Scope

Generative agents, LLM-driven social simulation, synthetic respondents and human digital twins, swarm prediction engines, and the infrastructure, research and money around them.

Out of scope: physical and engineering simulation, robotics sim-to-real, industrial digital twins, and general AI news that doesn't concern simulating people.

Pipeline

  1. 01
    Crawl

    Every 12 hours a scheduled job queries 14 sources: arXiv, Hugging Face Papers, Google News, GDELT, Hacker News, Reddit, GitHub, Bluesky, official company blogs and community event calendars.

  2. 02
    Deduplicate

    URLs are normalised (tracking parameters stripped, arXiv versions merged) and syndicated headlines collapsed, so one story appears once.

  3. 03
    Filter

    Each candidate is scored for topical relevance by an LLM editor (claude-opus-5-5) backed by keyword rules. Mentioning a tracked person isn't enough on its own — the text has to be about the field. Rejections are logged so they aren't re-judged every run.

  4. 04
    Verify

    Funding rounds and featured events are checked by hand against a cited source and marked Verified. Anything extracted automatically is labelled Auto-detected until someone confirms it.

  5. 05
    Publish

    The site is rebuilt from the new data and redeployed. Every item keeps a link to its original public source, the date the source gives, and the date we first saw it.

Authenticity rules

  • Nothing on this site is generated as content. Every item is a real public page, linked.
  • Where a one-line summary was written by a language model, it is labelled AI summary and is restricted to facts in the source text.
  • Dates are the publisher's own. If a page shows no date, we show when we first saw it and say so.
  • Deals and events carry Verified or Auto-detected so you know which is which.
  • Social accounts are only linked once we've confirmed they belong to the person.

What we can't see

  • LinkedIn has no public API for posts, and scraping it violates its terms, so LinkedIn posts appear only when the press covers them.
  • X (Twitter) requires paid API access; the integration is built and switches on when a key is configured.
  • Paywalled journals and private Slack/Discord communities are not crawled.

Corrections & suggestions

Spotted a wrong date, a missing deal, an off-topic item, or a person or company who belongs on the watchlist? Open an issue with a link to the source, and we'll fix it on the next run.The watchlist currently has 24 people and 26 organizations.