What did they actually say this month?
SOTA AI Exec Quote Check is a live agentic intelligence terminal. Pick a frontier-AI executive from the roster, the CRT dials out, and a structured briefing types back at you on what that person has said publicly in the last thirty days — sourced, paraphrased, and tonally read.
> SUBJECT........: [EXEC NAME] > ROLE...........: [ROLE @ ORG] > WINDOW.........: T-30D > SOURCES SCANNED: [N] [ KEY POSITIONS ] - [Paraphrased position, dated where available] - [Paraphrased position] - [Paraphrased position] [ NOTABLE QUOTE ] "[≤14-WORD VERBATIM QUOTE]" — [SOURCE], [DATE] [ TONAL READ ] [One sentence: bullish / cautious / defensive / cryptic this month?] [ SOURCES ] 1. [Title] — [Publication] — [YYYY-MM-DD] — [URL] 2. [Title] — [Publication] — [YYYY-MM-DD] — [URL] > BRIEFING COMPLETE.
Why a CRT terminal
The aesthetic is the product. Phosphor green on black. ANSI Shadow ASCII banner. Scanlines, modem dial-up on boot, mechanical keypress clicks while the briefing types in, chromatic aberration on the title block. Every interaction beat is composed for screenshot and screen-recording — this is publishable artifact tooling, not an admin panel. The CRT is the frame; the briefing is the payload.
The output format
Subject, role, T-30D window, count of sources scanned. Low signal stated plainly when sparse.
Three to five paraphrased bullets with dates where available. Not direct quotes.
One direct quote per briefing, fourteen words or fewer, attributed and dated. Never fabricated.
One sentence. Bullish, cautious, defensive, cryptic? The only editorial line in the briefing.
Numbered list with title, publication, date, URL. Drawn from the live web search.
Clipped, declarative, terminal-style. No emoji, no marketing register, no meta-commentary.
How it works
- Press any key to boot. Modem handshake, boot sequence, banner resolves out of chromatic aberration into clean phosphor.
- Tap a roster tile. Engage tone, screen flicker, secure-channel preamble.
- A server route fires
claude-sonnet-4-6with theweb_searchtool, scoped to the subject and the T-30D window. - The model returns the structured terminal template above. Stream comes back as Server-Sent Events; the client types it character-by-character with throttled keypress audio.
- On completion: copy briefing, share (deep-linkable to the same exec), or pick a new target.
Roster
Google DeepMind (4) — Demis Hassabis, Shane Legg, John Jumper, Koray Kavukcuoglu
OpenAI (3) — Sam Altman, Greg Brockman, Fidji Simo
SSI (3) — Ilya Sutskever, Daniel Gross, Daniel Levy
Thinking Machines (3) — Mira Murati, John Schulman, Barret Zoph
xAI (2) — Elon Musk, Igor Babuschkin
Interpreters (5) — Lex Fridman, Sean Carroll, Dwarkesh Patel, Tyler Cowen, Ezra Klein
Twenty-eight subjects across seven factions. Interpreters are not executives — they are the long-form podcast hosts and public intellectuals who interview the principals. Their briefings answer a different question: not what is this person building but what is this person currently sense-making about.
Rules of the briefing
- Paraphrase, do not reproduce long passages.
- Direct quotes under fifteen words; one quote per briefing, max.
- Never fabricate quotes or dates — if unsure, omit.
- If sources are sparse, the header says so plainly:
SOURCES SCANNED: 2 (low signal). - No editorialization beyond the single-sentence tonal read.
- Terminal voice. No emoji. No marketing register.
Where it ships
Same Vercel project, same route, both URLs aliased. Deep links of the form ?exec=<id> skip the boot sequence and auto-engage that subject — share links land directly on the briefing.
Lineage
A Context Jamming production by Bret Kerr. Sibling to Relational Intelligence / Thought Molecules RAG — the local agentic RAG over private deep-research corpora. Same operator, same allergy to fabricated quotes, different surface.
web_search tool, server-only API key, per-IP rate limiting. Audio synthesized or CC0. Ships at contextjamming.com/quote-check and quote-check.gemclaw.click.