Super Human Intelligence - A practical guide.

My bibliophile friends love to quote Umberto Eco: ‘At the age of 70, those who have never read will have lived only one life, their own. Those who read, on the other hand, will have lived 5,000 years’

I would like to amend this by adding a third element: those who use ChatGPT will have lived at least a million years. They can have access at any time to an extra brain that has read an infinite number of books. And together with the books they have read handouts, notes, critiques, popular science lectures and summaries.

The thesis of this guide is simple: we can criticise the limits of AI, we can denounce its dangers, of course, but on the other hand we have before us a great opportunity, which presents itself (almost) free for the taking: we can achieve Super Human Intelligence (SHI)

I don't want to be vague, on the contrary. I have tried to outline a very concrete and easy-to-implement structure that has all the elements to help the person starting out reach S.H.I. as quickly as possible.

Preliminary remarks: ChatGPT is not intelligent in itself.

It is a probabilistic system designed to create texts as similar as possible to those that a human with knowledge and medium intelligence could write.

We humans tend to anthropomorphise the things around us and therefore we say that ChatGPT reasons, thinks, responds, has hallucinations or gets it right, ‘gives me a hand’.

The truth is that it is a machine: without feelings, without inner purpose, without morals, without conscience, without knowledge. It is ‘stupidly’ limited. It does what it is told to do. Its ability is limited by the scope of the prompt it receives.

If I ask it to write me an email summarising a presentation I gave, it works; if I ask it to advise me on whether I've made the right life choice, clearly it doesn't work.

However, it has some great advantages: it has been trained on an enormous amount of texts, which includes - perhaps - everything that has been published so far.

In addition to this, it has the great advantage of being conversational. Given the right requests, it is able to combine this data to support our requests better than any search engine, fully understanding our writings and our arguments.

With these advantages in mind, here is the scheme I propose

Like it? help me grow: share this post.

Phase 1: Idea enlargement

The starting point can be a basic concept, a simple idea, a momentary intuition. The kind of thing that comes to mind while taking a shower or going for a solitary walk.

For example, the thesis of this article.

I then developed a series of questions that can help - from a momentary intuition - to derive a more articulated and solid structure of thought.

The process is based on ChatGPT. You can use the basic version. For convenience, I have built a custom assistant with instructions to answer me in a way that is more consistent with the style and tone I want. I named it Sokrates (like the famous Brazilian footballer).

You can test it here: 🏛️SOKRATES

The questions:

With our rough idea in mind, here is a series of questions I recommend asking:

‘What do you think of this reasoning... ? Help me rephrase this thought of mine in a more structured, clear and orderly way, with more academic vocabulary.’

Expected outcome: here ChatGPT rephrases the original idea. It usually trivialises a little. It's useful for understanding if the idea you want to develop has ground to grow on or is a rehash of a more trivial concept.

‘Help me structure my argument better. Can you tell me if there are any philosophers, writers or essayists who have dealt with this topic and whom I should quote? Who are they and what is their view on the matter?’

This question helps to place your idea in the existing debate. There may be (spoiler: there almost always are) writers or authors who have covered your topic before you and done a better job. It's worth giving it a read, exploring their ideas with different questions, and in the meantime understanding the similarities and differences between these people and your draft idea.

‘Try to make counter-objections to my idea. Criticise it. In doing so, if you can, quote texts and authors who have taken positions opposite to mine’

This is the most important question: usually chatgpt is very good at understanding the key points of an argument and dismantling them. It may be necessary to insist a little. The debate that is created is always very interesting

‘How would [author name] have approached this issue?’

This is a common method in design thinking exercises. It's also fun: it feels like you're having a dialogue with authors who died centuries ago.

Phase 2: Critical reworking

In phase one you should have come up with many ideas, some confused, some banal, but which have certainly strengthened your initial thought. You should also have come up with further reading suggestions.

The key thing here is to rework what you have collected.

Reworking means activities such as:

Writing down your thoughts
Keep a notebook with reading and study recommendations and fundamental notes on key concepts
Schematise ideas and counter-objections
Talk about them with friends and ask what they think
Observe the world and look for elements in real life that reinforce or deny the insights gathered through dialogue with the IA.

Phase 3 - writing and strengthening the model

Writing is an essential part of this process.

Firstly because writing is the visual reorganisation of a thought: it is one of the most effective methods for memorising ideas and concepts and for organising them in your head. It is a process that requires time and patience. It should be seen as an exercise: the more time you spend writing letters and diagrams on a sheet of paper, the easier it will be to write, reason and memorise in the future. Above all: the more order there is in your head, the better.

Secondly, written text is the preferred input for linguistic AI models, allowing them to further rework what has been written and give increasingly refined answers to what has already been thought and reasoned: in this way a sort of continuous dialogue is formed, which never starts from scratch, but continues for months, like possible university lectures.

For this phase I prefer paper (with its materiality and analogue timing) to writing on the computer. I then retype what I have written so that I can feed it to the AI.

Possible interactions with the AI

Lexical refinement

‘Can you rewrite this article in a more academic and a more popular style?’

Often in diary writing and in one's notes, colloquial, inexact, imprecise terms are used.

With this type of interaction, the AI suggests more precise, more fitting terminology. It is also an exercise to improve one's language in everyday life.

Words are also ideas, the more precise words we use, the more order and precision we will have in our thoughts.

Integration of the model with your own database of writings.

To do this it's useful to put all your writings on a public library. In my case I use notion. At this point in the back-end of 🏛️SOKRATES I have inserted a reference to my articles and in the prompt I have indicated

‘When you can and when required, refer to what Gabriele Sirtori has written in the articles that you can find at this link: [insert url]’

This is very useful because, at the end of a written piece, you can feed it to them and ask

‘Can you understand how what has just been written is consistent with the other texts written by Gabriele Sirtori?’

‘Which of Gabriele Sirtori's other writings do you think are related to this text I'm sending you, and how could they be cited as part of a single, more complex and organised train of thought?’

Phase 4 - Conclusion and new iterations

At this point, you start to structure a single, complex train of thought, made up of many aspects and many citations.

With time, and with new ideas that will gradually arise, this train of thought will become stronger and gain in uniqueness and robustness.

Above all, you will be able to form a second brain, trained and educated on your ideas, to which - when you feel confused - sometimes - you can try to ask: ‘What do you think, since you know what I think?’