Why a field guide, not a feed
AI news is abundant. What's scarce is a reliable answer to where does this actually stand? Feeds bury that answer under whatever shipped this morning. YLEM is organised the other way round: a small number of living pages, one per field, revised in place with a visible history.
Every piece of writing answers three questions: what changed, why it matters, and what you can build with it now.
The four parts
- Fields: the state of each area of the frontier, with stage, momentum and what to watch.
- Lab: hands-on tests with stated versions, environments, numbers and a verdict.
- Log: short dated notes. Observations land here first.
- Timeline and Index: the reference layer linking everything to labs, models and papers.
Lab verdicts
| Verdict | Meaning |
|---|---|
| Adopt | Use it in real work now. |
| Trial | Worth a real project, with care. |
| Watch | Promising, not ready for you yet. |
| Avoid | Not worth your time today. |
Verdicts are dated. AI tools change monthly, so a report always names the version it tested.
Field stages
Stages run from Research → Emerging → Maturing → Deployed. Momentum (accelerating, steady, slowing) is an editorial judgement about the pace of change over recent months, and is revised with the guide.
What we cover, and what we don't
We cover foundation models, reasoning, agents, AI coding, multimodal systems, memory and retrieval, robotics, compute, and alignment and safety, including the long-range question of superintelligence.
We don't cover celebrity AI stories, prompt collections, listicles of tools, or funding gossip. Nothing on YLEM is sponsored.
Corrections
Mistakes are fixed in place and recorded: field guides in their revision history, everything else with a correction entry in the log.
The name
Ylem is the term George Gamow and Ralph Alpher used for the hypothetical primordial substance from which the elements formed. The frontier of machine intelligence is still in that state: not yet settled into its final forms.