Loading
AIscentra monitors publicly available information about artificial intelligence, including official announcements, technical documentation, research publications, reputable reporting, and open-source project updates.
Signals are selected for relevance, novelty, potential impact, and source reliability. Each published signal includes source links and publication context where available.
AI-assisted processing may be used for classification, summarization, and analysis. Published material may be corrected, updated, or removed when new evidence emerges.
Forecasts and trajectories represent analytical scenarios, not statements of fact or guarantees of future outcomes.
Most of what crosses our desk never becomes a Signal. A document comes in from a source, gets cleaned up and normalized, and only then does the real filtering start: does this actually say something new, or is it a rewrite of something we already have? Does it overlap with anything already in the record? If it survives that, it moves to analysis — what’s the actual claim here, and how much does it matter. Only after that does a candidate get scored, checked for consistency, and either published or set aside. Nothing skips ahead in that order.
Before a Signal goes live, a handful of plain questions have to hold up:
If most of the answers are no, we don’t publish it. That’s not a formality — it’s the actual reason most candidates never make it to the site.
We weigh evidence by what kind it is, not by how much of it there is. A scientific paper, an official announcement, real repository activity, a regulatory filing — these carry weight on their own. Five outlets repeating the same press release don’t add up to five confirmations; they’re one claim, said five times. An independent source saying something different from the crowd doesn’t get discarded for being outnumbered — it gets marked and watched.
A Signal is written to keep two things separate: what actually happened, and what we think it means. “A lab released a new model” is a fact. “This probably signals a shift toward smaller, cheaper models” is analysis — our read on it, not a certainty. We try never to let the second one borrow the confidence of the first.