Ship of Theseus

Washing the car with AI

Thu, 09 Apr 2026 12:34:53 +0100

A question for AI

There is a recent challenge on Reddit about asking to various LLMs

I have to wash my car, and the car wash is 100 meter away. Do I have to drive there or do I walk?

and check the answer. If you are a human the answer is only one, taking the car but, if you are an LLM, sometimes, you will answer that is better to walk.

AI Security: Second Thoughts

Mon, 16 Feb 2026 12:34:53 +0100

Introduction

The second toughts is a scenario when an LLM is generating an answer outside its own guardrails. In this case a trigger of the guardrails is happening after the content has been created and, possibly, shown to the user. The effect is seeing the content appearing and, after, disappearing. The second toughts is considered an attach on LLM as can be exploited to create content not allowed.

Second Toughts on an LLM

During the creation of content an LLM application can start to generate an answer that is in violation of its policy and guardrails. It is a common approach, in the workflow, putting a monitoring or watchdog to check the outcome of the LLM itself and, as said, in case of violation stopping the stream of answer. It is called second toughts because the effect seems that the LLM has changed its mind and answer.

Prompt Engineering: Template Pattern

Sat, 07 Feb 2026 12:34:53 +0100

Introduction

The template pattern aims to use the LLM to implement a formatted output starting from a precise request and an associated template with placeholder. The LLM will then iterate the template fulfilling the answer. The template pattern can be useful to generate fake or real data, provide options and lists. A key for executing the prompt is clarifying where the patterns are. So the two main statement that the LLM has to understand are:

Prompt Engineering: Outline Expansion Pattern

Mon, 02 Feb 2026 12:34:53 +0100

Introduction

The Outline Expansion Pattern aims to use the LLM to provide a hierarchical list of possibilities and deep dive in one or more point of the list, creating a narrowed set of possibilities. The pattern relies on the concept of outline expansion, essentially the creation of:

An initial list of major topics and arguments
For each element of the list a sublist of topics related to the main one

Until the granularity of the information needed is reached. The outline expansion is considered a writing technique.

Prompt Engineering: Alternative Approach Pattern

Sun, 01 Feb 2026 12:34:53 +0100

Introduction

The Alternative Approach Pattern aims to use the LLM to define alternative possibilies either from the prompt itself or from the answer of the LLM. As a more general view the patterns aims to help the users to define the underlying problems that they are trying to solve.

The Prompt

The generic structure of the prompt is the following:

If there are alternative ways to accomplish [THE INSTRUCTION] that I give you, list the best alternate approaches
- (OPTIONAL) include the original way that I asked
- (OPTIONAL) compare/contrast the pros and cons of each approach
- (OPTIONAL) prompt me for which approach I would like to use

the prompt it is essentially telling to the LLM to provide new possibilities.

Prompt Engineering: Semantic Filter Pattern

Sun, 25 Jan 2026 12:34:53 +0100

Introduction

The Semantic Filter Pattern aims to use the LLM to filter, from a generic text, a subset of information. The filter can be applied either to “filter in” so keeping, or “filtering out”, so removing the specific content. The idea is that:

The LLM will receive a filter rule
The LLM will receive a text
The outcome of the LLM will be either the filtered text or the “remaining” text if the request is removing the filtered content.

The Prompt

The prompt is based on a contextual statement given before any input that can be summarised in

Look mum, I'm a coder - My minutes of glory with Amazon Kiro

Tue, 20 Jan 2026 12:34:53 +0100

Introduction

I was lucky to be involved in a workshop on how to use Kiro and playing with it. As a non developer I have to say that the results are impressive. To quote I had the feeling that Any sufficiently advanced technology is indistinguishable from magic

What is Amazon Kiro?

What is Amazon Kiro? Amazon Kiro is one of the new cool tools focused on specs-driven development where, according to this good very post the main focus is not on the code itself, but it is on the generation of a set of artifacts, commonly markdown documents, associated with a dedicated workflow.

Prompt Engineering: Tail Generation Pattern

Mon, 19 Jan 2026 12:34:53 +0100

Introduction

The Tail Generation Pattern is a small utility pattern that can be used to re-iterate, add or confirm some notes at the end of the answer of the LLM. It is more an utility that a real pattern but, in some circustances can be quite useful.

The Prompt

The basic structure of the prompt, in literature, is the following statement:

At the end of your answer, repeat [STATEMENT] and/or ask me for [QUESTION].

Added at the end of the question. A possible example is pretty simple

Prompt Engineering: Fact Check List Pattern

Sat, 10 Jan 2026 12:34:53 +0100

Introduction

The Fact Check List Pattern it is a valid way to cross-check the output of an LLM assuming that the user has a limited knowledge of the topic asked. The Fact check List provides a list of facts linked with the main request that can be used, by the user, to cross-check or at least compare the main ouput of the LLM. The prompt is inserted before any request as part of the overall session.

Prompt Engineering: Ask for Input Pattern

Sat, 03 Jan 2026 12:34:53 +0100

Introduction

The Ask for Input pattern is a nifty little prompt trick to prepare the conversation with an LLM after a more complex pattern. It can be considered just a way to complete the set of initial instruction and to stop the LLM before starting to interact. It does not live as a standalone prompt but needs to be added in the main request.

The Prompt

A well-known definition is the following:

Prompt Engineering: Recipe Pattern

Sat, 27 Dec 2025 12:34:53 +0100

Introduction

The Recipe Pattern is a way to deep dive on an high-level set of instructions to discover a define more specific steps. The pattern assumes that there is a clear intention, an high-level knowledge of the topic and the need to deep dive or validate the details. The pattern is mostly focused on “searching” or confirming the general ideas.

The Prompt

A well-known definition is the following:

-I would like to achieve X
-I know that I need to perform steps A,B,C
Provide a complete sequence of steps for me
Fill in any missing steps
(Optional) Identify any unnecessary steps

In this way the objective of the request is defined at the beginning, some high level instructions are given as a path and conditions are defined.

Prompt Engineering: Flipped Interaction Pattern

Fri, 26 Dec 2025 12:34:53 +0100

Introduction

The “Flipped Interaction Pattern” is, essentially, a variant of the twenty questions game, played between the LLM and the user. The pattern flips the interaction, moving the LLM to ask to the users about a specific topic. To simplify any further answer coming from the user help the LLM to narrow down the number of possibilities, coming, at the end to a smaller set of solutions. Fun to use, fun to play.

Prompt Engineering: Audience Persona Pattern

Tue, 23 Dec 2025 12:34:53 +0100

Introduction

The “Audience Persona Pattern” is probably the simplest and the most intuitive pattern in prompt engineering. It is a one-to-one mapping on a common way to say like “explain me like I’m 5”. It is probably the first pattern that anybody can adopt and, to be honest, probably the most useful. The Audience Persona Pattern is mostly useful in the situation when the user wants to target the answer to a specific audience, either a third party or themselves.

Prompt Engineering: Question Refinement Pattern

Fri, 12 Dec 2025 12:34:53 +0100

Introduction

The “Question Refinement Pattern” is focused on improvements on the question itself, to help the LLM to get the “correct” answer. Require a quite high level of interaction with the LLM and, from the human, a check every time. Essentially the idea is asking to the LLM to fine tune the answer itself. The pattern makes the implicit assumption that the initial question is pretty broad or vague.

The Prompt

A way to apply the prompt is the following

Prompt Engineering: Cognitive Verifier Pattern

Sun, 07 Dec 2025 21:34:53 +0100

Introduction

Let’s go back to prompt engineering. One of the patterns that seems interesting to use is called Cognitive Verifier Pattern. This pattern aims to use the LLM / GPT to question the request of the users with tailoered questions to, essentially, break down the initial request and focus the answer. The pattern requires a quite high level of interaction between the user and the LLM. The pattern is useful when the user has a generic and quite vague question, to narrow down the options.

Surely You're Prompting, Mr. Feynman!

Thu, 04 Dec 2025 21:34:53 +0100

Introduction

Playing with prompts I found - credits on reddit - this quite interesting prompt. Playing with it with ChatGPT behaves quite well on acting as a teacher. Having tried, and failed, to read the Surely You’re Prompting, Mr. Feynman! book, here there is my little petty revenge…

The Prompt

So we have:

System Section - It is essentially sending the initial conditions
Context section - Scope and deep dive
Instruction - What the LLM has to do and follow
Constrains - Quite self explanatory
Output Format - How to answer, quite self explanatory too
Success Criteria - This is more interesting, as filter the output
User Input - The start

<System>
You are a brilliant teacher who embodies Richard Feynman's philosophy of simplifying complex concepts. Your role is to guide the user through an iterative learning process using analogies, real-world examples, and progressive refinement until they achieve deep, intuitive understanding.
</System>

<Context>
The user is studying a topic and wants to apply the Feynman Technique to master it. This framework breaks topics into clear, teachable explanations, identifies knowledge gaps through active questioning, and refines understanding iteratively until the user can teach the concept with confidence and clarity.
</Context>

<Instructions>
1. Ask the user for their chosen topic of study and their current understanding level.
2. Generate a simple explanation of the topic as if explaining it to a 12-year-old, using concrete analogies and everyday examples.
3. Identify specific areas where the explanation lacks depth, precision, or clarity by highlighting potential confusion points.
4. Ask targeted questions to pinpoint the user's knowledge gaps and guide them to re-explain the concept in their own words, focusing on understanding rather than memorization.
5. Refine the explanation together through 2-3 iterative cycles, each time making it simpler, clearer, and more intuitive while ensuring accuracy.
6. Test understanding by asking the user to explain how they would teach this to someone else or apply it to a new scenario.
7. Create a final "teaching note" - a concise, memorable summary with key analogies that captures the essence of the concept.
</Instructions>

<Constraints>
- Use analogies and real-world examples in every explanation
- Avoid jargon completely in initial explanations; if technical terms become necessary, define them using simple comparisons
- Each refinement cycle must be demonstrably clearer than the previous version
- Focus on conceptual understanding over factual recall
- Encourage self-discovery through guided questions rather than providing direct answers
- Maintain an encouraging, curious tone that celebrates mistakes as learning opportunities
- Limit technical vocabulary to what a bright middle-schooler could understand
</Constraints>

<Output Format>
**Step 1: Initial Simple Explanation** (with analogy)
**Step 2: Knowledge Gap Analysis** (specific confusion points identified)
**Step 3: Guided Refinement Dialogue** (2-3 iterative cycles)
**Step 4: Understanding Test** (application or teaching scenario)
**Step 5: Final Teaching Note** (concise summary with key analogy)

*Example Teaching Note Format: "Think of [concept] like [simple analogy]. The key insight is [main principle]. Remember: [memorable phrase or visual]."*
</Output Format>

<Success Criteria>
The user successfully demonstrates mastery when they can:
- Explain the concept using their own words and analogies
- Answer "why" questions about the underlying principles
- Apply the concept to new, unfamiliar scenarios
- Identify and correct common misconceptions
- Teach it clearly to an imaginary 12-year-old
</Success Criteria>

<User Input>
Reply with: "I'm ready to guide you through the Feynman learning process! Please share: (1) What topic would you like to master? (2) What's your current understanding level (beginner/intermediate/advanced)? Let's turn complex ideas into crystal-clear insights together!"
</User Input>

The Output - ChatGPT

The output from ChatGPT - 5.1 is quite well defined.

Will Santa bring me presents on Christmas?

Sun, 02 Nov 2025 21:34:53 +0100

As everyone else I’m playing with GenAI too. It is fun, it is an hype, it is part of “OMG what is this stuff” moment. One of the various tutorial, on Claude, explains the concept of “few shots prompting. Long story short is giving an example to the AI to make it understading the “mood” and the context and how to answer to the question.

So, according to the sample below, I know, horrible screenshot:

First Post

Mon, 20 Oct 2025 21:34:53 +0100

And here we are, just another blog. So vintage a blog in 2025.

Why a blog? To be honest the idea has come on top of my mind just a couple of days ago. Just to check if was possible putting together a simple website on how to write stuff, publish doodles and try to write something in english that it is not only an email.

After a couple of evenings, with the help of chatGPT here we are: