We found no evidence of formal reasoning in language models

CEOs, researchers, heads of state, senators, philosophers, essayists, anonymous engineers, marketing departments, and the models themselves. Every quote in the archive links to its source, and wherever it can be found, that is the original: the speaker’s own post or essay, the interview, the transcript, the outlet that first reported it. Never a copy of a copy. Quotes are verbatim, shortened only by cutting at the edges, never rewritten.

Two ways in. Some quotes are chosen by hand and meant to hold for years, though the selection does change: the classics, the slogans, the vocabulary, and a few lines we wrote ourselves, always marked as ours. Most quotes from before 2022 came later: a model found them in primary sources, and the code checked each one word for word against the full text.

Most arrive with the news: they are distilled once a day from 170 sources: media, company blogs, newsletters, philosophy magazines and research institutes, plus news searches for the people and topics that shape the field, from lab CEOs and heads of state to the philosophers asking what any of this means (the full list), by a language model working under fixed rules, then checked in code: length, no names on the wall, no duplicates, and for quotes the exact wording must appear in the source. A person reads every proposal before it goes live at noon and can remove it. Every quote is then checked a second time against its source: the exact words, the speaker, the date. What can't be confirmed doesn't go up. No voice gets more than one line a day and no outlet more than three. Quotes from the months between November 2022 and the start of the daily update were added later the same way, from the news of their day, and reviewed one by one.

There is no daily quota: a quiet day adds nothing, a loud one adds a dozen. News quotes leave the wall after 30 days and stay in the archive. When no one in particular said a line, it says what it is instead of a name: a company’s slogan, a document, a term, words an AI produced, a line of our own.

Every quote is verbatim unless marked as a paraphrase, dated and linked to where it was said, from Ada Lovelace in 1843 to this morning’s news, and the wall adds no commentary and no ranking. Each quote carries a line of context and a tab “Who else?” with other voices on the same point; you decide who agrees and who pushes back.

From any quote, the speaker, the topic and the year are one click away, and every speaker page shows where others mention that person. The whole set is open data under CC BY 4.0.

Partly. The daily quotes are proposed by Claude, an AI by Anthropic, and filtered by code; a person reads every proposal before it goes live and can remove it. The hand-picked quotes, our own lines and everything on this page are written by people. Which is to say: the wall is about AI, made with AI, and the arguments on it are not the machine’s.

Once a day at noon Berlin time. The ring around the date fills over 24 hours; click it for the countdown. In the archive, a new quote shows its date in green with a green dot for a day. Click the date to see everything new today.

The note at the bottom of the wall counts everyone who has it open right now, one per device, refreshed every 15 seconds. No names, no accounts, nothing stored.

• Click the smiley: one small smiley per person jumps out
• Throw them around: stack them on the quote

The Like button in the quote view and in the archive saves the quote in your browser, nowhere else. The 3 button at the top of the wall then shows only your liked quotes on the wall. Clear your browser data and your likes are gone.

Each like also adds one to a public count per quote. It is anonymous: no account, no name, no IP address stored. While you have the wall open and someone else likes a quote, a short note at the top tells you which one and links to it:

Yes. AI Zeitgeist is a free plugin for TRMNL, the e-ink display for your desk or wall. It shows a new quote every hour, with speaker, source and a QR code to the full quote. Choose light or dark in the plugin settings.

Add to TRMNL

The quotes belong to whoever said them; short quotes with attribution are fine under quotation rules. The wallpaper download and the embed code are for personal use. For anything else, write.

Büro Bloock, a design studio in Cologne, Germany. AI Zeitgeist is an independent project, not funded by anyone on the wall. Errors, missing sources, a quote that should not be here: hello@aizeitgeist.tv.

Yes. The press page has a short profile in English and German, key facts, two videos (the launch reel and a how-to reel), logos as SVG, EPS and PNG, and 29 screenshots of the wall, the archive, the speaker, topic and year pages, the mobile view and TRMNL, all in one ZIP. The compilation is open data under CC BY 4.0; the quotes stay with their authors.

Open the Press Page

Yes. We won’t say where, that would spoil it. It is off the record, it takes a little persistence, and so far it has mostly been found by people who click things they probably shouldn’t. If you find it, keep it to yourself. Or don’t: everything on this wall ends up quoted eventually.

No ads, no profiles. Fonts are served from this site and your settings live in your browser. Likes are counted anonymously: one number per quote, nothing about you. We use Google Analytics to see which quotes travel, but only if you allow it when we ask on your first visit. You can change that choice any time. See the privacy notice and imprint.