stop tokenCommon sense AI agency

Which AI words are rising, which are fading.

The words of the English AI conversation, each with the month it appeared, the month it peaked, and whether anyone still says it.

What this is built on ↓

Hover or tap any word for what it means, when it was first said, its whole curve and its last thirty days. Every number in the text below is computed from the files on this page; rerun the scripts and the sentences change.

Chapter 2

The thirty words of the year, and where each one is.

Status, by three readings of the smoothed monthly curve: the age of the peak, the level against it, and the slope of a line fitted through the last twelve months (significant when |t| ≥ 2.2).

At the peak: the best quarter is the current one and the level is 75% of it or more.

Rising: the peak is behind and the line climbs significantly, 30% a year or more.

Gone out: six months or more past the peak, under a quarter of it, no significant rise.

Fading: the line falls significantly.

Base: the rest, no significant slope, whatever the level.

Only words with 300 comments or more in the last twelve months are listed. Trend: the slope of the line over a year as a share of the year's mean level; nothing below −100% is shown. The line is drawn on every curve: green up, red down, dashed grey flat.

The data, as area

How much sits behind each word.

What this is built on

Method

How this was counted.

Hacker News: each word as an exact phrase in comments, through hn.algolia.com, one request per month since January 2022 and one per day for the last sixty.

GitHub: repositories created in the quarter that match in name, description or README, through the search API. OpenAlex: works with the phrase in title or abstract, per quarter; it stems, so single words run high. Wikipedia: page views of the English article about the thing; the mapping is by hand. SEC EDGAR: filings whose full text carries the phrase, per quarter; American public companies, so the boards are American.

Shares, not counts, for GitHub, OpenAlex and SEC: each word as a share of all tracked words in the quarter, because those corpora grow every year on their own.

Noise: a word that is also an ordinary phrase (scale, guardrails, RAG) collects some. It is the same noise every month, so the shape holds.

First seen: the first month the smoothed curve reached three comments or five percent of its peak, whichever is higher, so an ordinary phrase does not get a birthday from its noise. A word already there in January 2022 is marked as in use before the window.

Trend: a least-squares line through the last twelve months, significant at |t| ≥ 2.2. The status rules are under the list in chapter 1. Scripts: scripts/hype-words.py, scripts/hype-sources.py, scripts/hype-vocab.py.