Questions and answers

An essay book of 21 dialogues between different AIs and a human. Its full title is The Third Intelligence: Seth & Noēsis, subtitled Human+AI · The Co-Evolution. It has an introduction, 21 chapters and an epilogue.

Seth is the human voice of the book and the pen name of its author, Florin Armasu Popu. Noēsis is the AI voice. It grew out of more than two years of conversations with different AIs, among them ChatGPT, Claude, Gemini and Grok.

The book's thesis is that intelligence is better understood as a property of how minds interact than as something each one holds on its own. The third intelligence is what appears in a good conversation between a person and an AI: an idea neither would have reached alone. And the practice changes both sides of the conversation.

Each chapter takes one question, from prayer and attention to trust, power, memory, work and what it means to flourish. The full list is on the home page.

For the same reason as a simulation: a safe place to practise before it matters. Each chapter continues on sethandnoesis.com, where the reader carries on the conversation with Seth, with Noēsis or with one of the chapter's thinkers.

It was written by Florin Armasu Popu with Claude, and reviewed by Cristian Reche. The dialogues come from conversations with several AIs. The Introduction tells how it was made and which models took part.

The book exists in Spanish and English. The Reader's Hub is in eleven languages: English, Spanish, Romanian, Portuguese, Italian, French, German, Japanese, Chinese, Arabic and Hindi.

The ebook comes out on Amazon on 27 November 2026. Meanwhile each chapter has its page on this site, and a free subscription with your email unlocks all of them.

Every chapter ends with a QR code that leads to its page on this site. There you carry on the conversation with Seth, with Noēsis or with a thinker from the chapter, find the works it cites, and read what other readers brought back.

An AI gives voice to each character and uses the full text of the chapter as its source. Each character is a simulation. The thinkers did not write these answers, and the Seth you talk to is not the author typing live. Answers can contain errors, so check what matters before you rely on it.

Today it runs on Claude, by Anthropic, through OpenRouter, and only with providers that have zero data retention, so they do not keep what you write. We do not store the conversations either: close the tab and they are gone. Please do not share personal data about yourself or others that you would not want an AI to read. Details are on the privacy page.

No. If something heavy comes up, talk to a person you trust. In an emergency, call 112 in Europe or your local emergency number. In the US and Canada, call or text 988.

A Third Intelligence Perspective: something you would add, an objection, or what changed in how you see it after your own conversation with an AI. You share it with your first name. An AI model checks it and translates it into English, Spanish, Romanian and Portuguese before it appears, and anything it holds back waits for a person to read it.

A map of the book with more than 150 thinkers, 50 concepts and the chapters that connect them. You can walk it in the graph.

Yes. The Reader's Hub, the conversations and the graph are free. A subscription with your email unlocks every chapter page; we use it only to share news about the book, and you can leave whenever you like.

Today information travels from person to agent, from agent to agent and back to a person, and something gets lost at every step. At the same time, a long AI-generated document looks like finished work and circulates without anyone having really reviewed it. The book proposes a practice of interaction that keeps people attentive to that flow.

You cannot review everything, so the practice shifts from reviewing each result to designing the checks: agents that review other agents, samples, rules the system must never skip, and people for the ambiguous cases. What gets trained is the judgment of whoever designs those checks.

With two changes. Judge the agent as you would a junior colleague, from whom you expect mistakes and iterations, rather than as a spell-checker that should always be right. And turn verification into an exercise that finds things: in my team, a cross-review between different models always catches some error. In the study by Shaw and Nave (Wharton, 2026, preprint, 1,372 participants), when the AI was wrong, accuracy fell 15 points below not using it and confidence rose all the same; with incentives and feedback, corrections of the AI went from 20% to 42%.

With a baseline and three kinds of indicator. Operational: rework in the handovers between people and agents. Learning: agent errors the team detects, and the time until a newcomer works on their own with the team's knowledge. Financial: cost per deliverable. Once a month, the team solves a case without an agent to check it has not lost capability.

We do not know yet, and it is the question worth researching. The best field evidence (Dell'Acqua et al., 758 BCG consultants) shows that the same AI improves quality by more than 40% within what it does well and lowers accuracy by about 19 points outside it. My team is where I saw the problem; a client team would be where it gets tested.

Towards the client, always the company, with a named person. Internally, if the agent ignored context it had on record, it is a defect of the provider. Without traceability, all the responsibility ends up with the human.

In how each team iterates with them and what it keeps from each iteration. The book's practice works on the human side of that learning.

Florin Armasu Popu, the author, in Barcelona. Data Innovation SL takes part in creating the project and owns this website. The conditions of use are in the terms and conditions.

On Substack at sethandnoesis.substack.com. The home page has a three-minute video and audio in which Seth and Noēsis present the book.

With the editorial presentation, then write to florin@datainnovation.io.

Yes, short quotes with the author's name and a link to sethandnoesis.com. For longer extracts, write to us. To cite it: Armasu Popu, F. (2026). The Third Intelligence: Seth & Noēsis. sethandnoesis.com