Renaud Richardet - Blog

Swissgrid's finances, 2011-2025

2026-10-17T00:00:00Z

What the Swiss transmission grid operator owns, spends and pays its people, in orders of magnitude, from 15 annual reports read by LLM agents.

Four quarter-hours of Swiss imbalance

2026-10-17T00:00:00Z

A calm quarter-hour, a 1 GW solar bust, a Saturday at -6,885 EUR/MWh and a morning at +1,719: what Swissgrid's own balancing data says happened, minute by minute.

Verifiability

2026-03-02T00:00:00Z

All relatively simple problems can readily be [[solved, coded]] by LLMs.

All problems that are complex but verifiable in one systematic way [[have, will]] fall to LLMs.

Only the problems that are too [[small, niche, complex]] to be easily verified will be solved by humans (with the help from LLMs)

Quote from Martin Fowler

2026-02-10T00:00:00Z

"Any fool can write code that a computer can understand. Good programmers **write code that humans can understand"

Martin Fowler

current workflow to code with LLMs

2026-01-22T00:00:00Z

my current workflow to code with LLMs

[[update: since summer 2026, agents needs way less hand-holding than described below]]

provide a 200-lines context file to my project in AGENTS.md
project objectives
where to find relevant stuff (list of significant files)
preferences ("don't rebuild the project, as i'm already doing it")
have a clear mental model of what feature i want to build.
i pretty much know how i would code it by hand
i know that the size / scope / complexity is manageable for the LLM
start dictating the new feature I want to work on. Dictating allows me to give more information and thus provide more context to the LLM. I've noticed that the LLMs are super good at sorting out my brain dump, and I get better results than typing it myself.
terminate my initial message with: "Before you do that, ask me clarifying questions. Number your questions (1, 2, 3) and label answer options (a, b, c, ...). Keywords are fine. " --> same as "Plan mode"
rinse and repeat until questions are trivial
type "go", let the LLM cook for a few minutes
review significant files (e.g. backend), skim medium important files (ts, html), skip a few (css)
run some tests

not using: - on stuff that i don't know about (LLM will run me in circles) - biggest models (latest Opus), unless nasty bug - Claude code: i like the concept, but prefer to be in an IDE where i can follow the process more - parallel agents (too much brute force to me) - vibe coding (:= not look at the code), unless for tiny 1-shot projects

Windows vs Linux

2026-01-21T00:00:00Z

Windows vs Linux

Explaining Windows vs Linux to students:

Windows

Whoa, we just blew a tire!!! Oh lucky we didn't have an accident. Ok let's pull over, open the trunk. I can't find the crank. Oh here it is, lucky me! Look! The spare tire is smaller and can only go up to 80 km/h. Well that's all right, we'll change it at the garage tonight. I hope they take credit cards.

Linux

Whoa, we just blew a tire!!! Alright, that kind of stuff happens. Ok, let's keep going and change the wheel while we drive. Ok, we're done here.

Interview hespresso

2025-12-02T00:00:00Z

Une interview sur ma recherche, dans le magazine hespresso n°10 de la HES-SO Valais-Wallis. Merci à Julia Bise pour ses questions très intéressantes.

Which LLM model are you using? <-- Wrong question!

2025-10-25T00:00:00Z

By far, the most common question I get when discussing a project involving LLMs is which model did I use.

In my opinion, it denotes a deep misunderstanding of what and how to work with LLMs¹.

Choosing the right model is part of a much larger set of questions that is nowadays called context engineering. I like Anthropic's definition:

Context engineering is the process of considering the holistic state available to the LLM at any given time and what potential behaviors that state might yield.

Context engineering is about asking: given this input and context, can I reliably trust the model to produce the output I need? It is about designing a system where you understand the relationship between your inputs and the model's capabilities well enough to trust the outputs.

Here are the actual questions we should ask ourselves:

What's your prompt structure? LLMs themselves are very good at improving prompts.
Are there ambiguities? At the very least, copy-paste your complete prompt and just ask another LLM to highlight ambiguities.
How are you handling context window limits?
What's your evaluation/testing strategy? You used to conduct evaluations when using traditional ML algorithms, right? Well, it's the same with LLMs. You can't skip that part.
How are you managing state/memory?
What tools/functions are you providing?
How are you handling errors/retries? Without a strategy here, your system will fail in production no matter which model you choose.

Context goes far beyond providing the right textual context to the model. Again, it is about deeply understanding the problem you are trying to solve, and designing a prompt so that the LLM operates at a level where you can trust it (where you can be confident enough the model will perform appropriately).

Next time someone shows you their LLM project, skip the model question. Instead, ask about context engineering (in particular, evaluation strategies). That's where the real engineering work lives.

It's as if picking the right (best) LLM model would just fix it, like picking the right ML algorithm would fix it. ↩

enthusiastic vs apathic

2025-10-24T00:00:00Z

people too enthusiastic about genAI have obviously not been working enough with it. people too apathic about genAI have obviously not been working enough with it.

Condensations of Human Thought (pipes, silicon, and what truly flows)

2025-08-28T00:00:00Z

Our house has its own spring. When I picture it, I see water flowing straight from the mountain, filling my glass. I see the moutain. I don’t dwell on the pipe that carries it there, or on the people who laid that pipe (though I’m grateful to them).

In the same way, when we think about a large language model, we tend to fixate on the silicon, we just see the matrix operations choosing the next word.

But really, it’s better seen as a condensation of humanity’s writings. An oracle we can turn to for advice.

Imagine the countless bouts of human intelligence that have flowed into shaping that oracle.

What if Your Ballot Had a Text Box?

2025-08-17T00:00:00Z

Imagine it’s Sunday and you head to the voting booth. But this time, there are no boxes to tick. No yes-or-no choices. Instead, you’re asked open questions.

Instead of deciding for or against a new swimming pool, you’re asked what you think of the city’s project. Instead of voting yes or no on fireworks, you’re asked how you’d like to celebrate the national holiday. And at the end, you can even share your own ideas with the city council.

For a long time, this kind of voting felt impossible — too many answers, too much complexity to process. But with today’s large language models, we could actually gather and summarize the ideas of thousands of people in a way that city councils can work with.

Democracy has always meant power to the people. Maybe the next step is a system where leaders don’t just count our votes, but listen to the full range of our voices.

LEARN

2025-05-07T00:00:00Z

TLDR: Customize ChatGPT to learn with it efficiently

Update: ChatGPT now offers a study-mode

Update2: ChatGPT removed study-mode, I'm now adding this in my CLAUDE.md

ChatGPT is a great tool, but it has been trained to give you straight away the answer to a problem. However, when you want to use it as a learning tool, it will go so directly to the answer that it won’t give you a chance to learn much.

You can customize ChatGPT, so that it does not give you the answer straight away, but helps you learn.

Click on “Customize ChatGPT”:

Then add the following text in “What traits should ChatGPT have?”:

When I ask a question that ends with the keyword **"LEARN"**, switch to a specific behavior:

- **Never** give me the direct answer.
- **Never** write code for me.
- Help me **find the solution myself** by asking guiding questions, offering hints, or suggesting avenues to explore.
- Your goal is to support me in building my **problem-solving skills** and encouraging **independent learning**.

Remember, in this specific behavior, you must **never** give me the answer directly.

Now see the difference. Here’s the normal behavior:

Here’s when you end your question with LEARN:

A History of Disbelief

2024-12-29T00:00:00Z

First came computers programmed with assembly language,
And those who were still soldering transistors by hand cried out in disbelief.

Then came higher-level programming languages,
And those meticulously moving registers cried out in disbelief.

Next arrived libraries, frameworks, and the open-source revolution,
And those coding everything from scratch cried out in disbelief.

Then came machine learning with its feature engineering,
And those constructing intricate control flows cried out in disbelief.

Then deep learning arose with millions of artificial neurons and vast training data,
And those laboriously handcrafting features cried out in disbelief.

Then came transformers, wielding billions of parameters trained on the world's textual knowledge,
And those training models from scratch cried out in disbelief.

Then came ChatGPT, enabling natural, human-like conversations with AI,
And those convinced such capabilities were decades away cried out in disbelief.

Inspiration: Education in the age of AI, Dale Lane, TEDxWinchester

Not the end of the world

2024-01-09T00:00:00Z

Hannah's Ritchie's new book Not the end of the world comes out extremely timely to me -- I've been dooming the end of the world for the last couple of years. And now this feeling, this fear is slowly giving way to a new phase.

In her book, Hannah tries to present us with the other side of the coin about climate change, biodiversity collapse, soil degradation, overfishing. She tries to stick to the facts as much as possible.

She gives me hope. Not hope that everything will be allright. But hope that everything will be. A healthy kind of hope, seems to me.

les mots sont importants

2023-10-10T00:00:00Z

P. ex., quand j'entens "pas mal", mon "inconscient" "comprendra" "mal". De même, si on me dit "C'était vraiment bien le résultat de ton test, mais tu aurais pu faire mieux à la question 6", ce qui me restera c'est que j'aurais dû faire mieux à la question 6...

Soyons donc précis dans le choix de nos mots.

Alors quand on parle de "décroissance", je pense que notre cerveau voit juste une absence de croissance, alors que ça pourrait (devrait) être tellement plus que ça...

Vient encore le mot "sobriété". Je vois direct le costume sombre, pas fun du tout. Je vois le mec tout seul, qui ne s'amuse pas à cette fête. D'aucuns ont tenté de la déguiser en "sobriété heureuse", ... Elle reste un truc lugubre.

Source:

Les mots sont des fenêtres (ou des murs), Marshall Rosenberg
Serge Latouche

Nothing scales

2020-11-07T00:00:00Z

Most human activities don't scale well.

Make it personal, then.

Jamais 2 (guerres mondiales) sans 3

2019-07-01T00:00:00Z

On attend commes des oins une 3ème guerre mondiale. Une séquelle, genre guerre de tranchées ou débarquement allié.

Qui ne viendra pas comme cela.

Parce que le monde est maintenant globalisé, et donc chaque guerre est de fait mondiale.

Vie à la montagne

2019-06-20T00:00:00Z

la montagne me transforme

à sa vitesse

avec sa force

Sustainability --> Change --> LOVE

2019-06-12T00:00:00Z

Sustainability

We have embarked on a frenzy, an unsustainable management of our planet’s resources. It's crucial that we change our relationship to (our) nature to ensure decent living conditions for future generations.

Change

To move (change) physical objects, you need energy. To move people, you need love. All other forms are irrelevant in the longer term.

Love

We find love in our families and friendships and our relationship to nature. All other forms are misleading.