---
sheet_format: 1.0
id: 2024-09-maci-116
title: "Les IA conférencières recentrent Voyager 1 en CSS"
title_en: "Conference-going AIs re-centre Voyager 1 in CSS"
speaker: Erwan Rougeux
with: [Horacio Gonzalez (host), Valentin Maerten, Gilles Biannic (Clever Cloud)]
date: 2024-09-27
event: "Message à caractère informatique #116"
place: "online / studio"
format: podcast
language: fr
duration: "57:07"
video: https://www.youtube.com/watch?v=_nSE1CFa_50
slides: null
topics: [tech conferences, digital sovereignty, africa, ai-assisted coding, rust, chatgpt, ai ecology, voyager 1]
source: "Summary written by Claude (AI) from the local Whisper transcript, with statements attributed to Erwan by local identification of his voice only (other speakers are not identified). Not a verbatim transcript; check the recording before quoting."
review: spot-checked
review_date: 2026-10-18
---

# Les IA conférencières recentrent Voyager 1 en CSS

**In one sentence:** A MACI episode on tech conferences, CSS centring, Voyager 1 and AI, where Erwan calls on public authorities to back developer events, recounts building a handpan tablature web app in Rust with ChatGPT, and argues for using AI where it really adds value.

## TL;DR
Host Horacio Gonzalez, Valentin Maerten, Gilles Biannic and Erwan Rougeux start with Touraine Tech, which Valentin co-organises, and how regional conferences live on sponsorship. Erwan, back from a mission in Abidjan, contrasts French public authorities with Côte d'Ivoire, where ministries invest in tech events. The hosts then cover CSS `align-content` for vertical centring and NASA's risky thruster manoeuvre on Voyager 1. The AI segment compares everyday uses (Cursor, Mistral, GitHub Copilot). Erwan tells how, over a few late nights, ChatGPT helped him turn a niche need (handpan tablatures) into an open-source Rust web app hosted on Clever Cloud, and ends with warnings on energy use, respect towards AI and narrowed information.

## Erwan's key points
- He says he has met extraordinary people at trade shows and events who shared their passion; the internet is fabulous, but nothing replaces meeting in person ([04:46](https://www.youtube.com/watch?v=_nSE1CFa_50&t=286s)).
- He is surprised that public authorities do not see the role of tech conferences in the economy: they are professionals and experts sharing knowledge, not "a nice meet-up of geeks", and he asks public services to support this ecosystem ([05:17](https://www.youtube.com/watch?v=_nSE1CFa_50&t=317s), [06:12](https://www.youtube.com/watch?v=_nSE1CFa_50&t=372s)).
- Back from a mission in Abidjan with Quentin Adam, he reports that in Côte d'Ivoire digital-transition ministries invest heavily and take part in tech events. He relays Quentin Adam's point that the United States hold about 80% of the global tech market and France about 0.4%, and cites the US Small Business Act, with no European equivalent ([08:31](https://www.youtube.com/watch?v=_nSE1CFa_50&t=511s), [09:27](https://www.youtube.com/watch?v=_nSE1CFa_50&t=567s), [09:51](https://www.youtube.com/watch?v=_nSE1CFa_50&t=591s)).
- He spends much of his working life in a suit and tie to open doors, and argues that these subjects must be settled by several actors around a table, with politicians from all sides invited to tech events, building a bridge between technical practice and its economic and political consequences ([10:44](https://www.youtube.com/watch?v=_nSE1CFa_50&t=644s), [11:14](https://www.youtube.com/watch?v=_nSE1CFa_50&t=674s)).
- On Voyager 1, more than 25 billion km away, he admires the technical feat and the humanist reach of a project linked to Carl Sagan, kept alive with the help of retired NASA staff and their old notes. Science fiction such as Star Trek inspires and guides humanity ([20:33](https://www.youtube.com/watch?v=_nSE1CFa_50&t=1233s), [22:38](https://www.youtube.com/watch?v=_nSE1CFa_50&t=1358s)).
- He uses Llama 3.1 most, because it runs locally. To test ChatGPT on a niche subject, he asked it for handpan tablatures, then for a Rust web page: it proposed parsing compressed MusicXML scores, drawing the handpan in SVG, MIDI synthesis and transposing 88 piano notes onto the instrument's 9 to 12 notes ([30:10](https://www.youtube.com/watch?v=_nSE1CFa_50&t=1810s), [33:25](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2005s), [34:09](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2049s)).
- He then had ChatGPT split, comment and document the code with cargo before hosting it on Clever Cloud; it flagged a zip-bomb risk in the uploaded files. About 150 exchanges and a few nights took him from idea to product. He sees AI as a catalyst for start-ups without an R&D team and for countries with technology gaps ([34:51](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2091s), [35:58](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2158s), [37:03](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2223s)).
- He learned to code with Stack Overflow, "copy, paste, adapt", and used ChatGPT the same way. He designs algorithms with pencil and paper; the model just translates an idea and a need into a programming language ([38:16](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2296s), [38:38](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2318s), [40:26](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2426s)).
- With only 8 GB of GPU, a local model takes about a minute to a minute and a half per answer against about 15 seconds for ChatGPT. He gives his AIs first names and stays polite with them, to keep our social codes. Because ChatGPT sessions can be lost, he asks the model to write a prompt that would restore its context; his assistant "Karl" now does so every evening ([45:06](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2706s), [47:18](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2838s), [47:40](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2860s)).
- He argues against using these tools as improved search engines, which burns GPU for little; asking for a program that then runs on an efficient host makes more sense. Like blockchain, not every project needs AI ([48:25](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2905s), [49:30](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2970s), [49:52](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2992s)).
- He fears AI answers act like blinkers: one compiled answer instead of several pages read with your own judgement, which could produce generations with a narrow view of complex subjects ([54:36](https://www.youtube.com/watch?v=_nSE1CFa_50&t=3276s), [54:58](https://www.youtube.com/watch?v=_nSE1CFa_50&t=3298s)).

## Outline
- [00:00](https://www.youtube.com/watch?v=_nSE1CFa_50&t=0s) Opening of episode 116.
- [00:31](https://www.youtube.com/watch?v=_nSE1CFa_50&t=31s) Introductions: Gilles Biannic (Clever Cloud, Brest, VM deployment team), Erwan (international growth at Clever Cloud), Valentin Maerten (Tours, Pubstack, Touraine Tech, Taskfile), host Horacio Gonzalez.
- [02:09](https://www.youtube.com/watch?v=_nSE1CFa_50&t=129s) Valentin on Touraine Tech: a deliberately family-size conference; CFP open until 31 October; 6–7 February 2025; sponsors wanted.
- [04:46](https://www.youtube.com/watch?v=_nSE1CFa_50&t=286s) Erwan: human encounters, and a call to public authorities to support tech events.
- [06:12](https://www.youtube.com/watch?v=_nSE1CFa_50&t=372s) Valentin: the university hosts the event; a two-day ticket at 50 euros, so sponsors fund most of it.
- [07:37](https://www.youtube.com/watch?v=_nSE1CFa_50&t=457s) The host recalls Quentin Adam's keynote at JUG Summer Camp: geeks should speak the language of public authorities.
- [08:31](https://www.youtube.com/watch?v=_nSE1CFa_50&t=511s) Erwan back from Abidjan: Côte d'Ivoire's ministries invest in tech; market-share figures; the Small Business Act.
- [12:35](https://www.youtube.com/watch?v=_nSE1CFa_50&t=755s) The host on CSS `align-content`: vertical centring without flexbox, in 2024.
- [17:03](https://www.youtube.com/watch?v=_nSE1CFa_50&t=1023s) Gilles: NASA restarts an old Voyager 1 thruster to keep the probe pointed at Earth.
- [20:07](https://www.youtube.com/watch?v=_nSE1CFa_50&t=1207s) Erwan on the distance, Carl Sagan and retired NASA staff.
- [23:48](https://www.youtube.com/watch?v=_nSE1CFa_50&t=1428s) Erwan: AI is presented as a threat to jobs, while its value as an accelerator in many countries goes unseen.
- [24:58](https://www.youtube.com/watch?v=_nSE1CFa_50&t=1498s) Valentin on Cursor, an AI-first editor; its pricing.
- [27:14](https://www.youtube.com/watch?v=_nSE1CFa_50&t=1634s) Gilles uses Mistral as a natural-language search engine.
- [28:22](https://www.youtube.com/watch?v=_nSE1CFa_50&t=1702s) Valentin: AI to explain code, GitHub Copilot and Inline Chat in IntelliJ as an accelerator.
- [30:10](https://www.youtube.com/watch?v=_nSE1CFa_50&t=1810s) Erwan: local Llama 3.1, then a handpan experiment with ChatGPT.
- [32:42](https://www.youtube.com/watch?v=_nSE1CFa_50&t=1962s) From ASCII tablatures to a Rust web app: MusicXML, SVG, MIDI, transposition.
- [34:30](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2070s) Open-sourcing it: refactoring, documentation, security review, zip bombs, hosting on Clever Cloud.
- [37:47](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2267s) Does it help people learn to code? Erwan on Stack Overflow and copy-paste-adapt.
- [42:35](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2555s) A generic model (GPT-4o mini) went further than he expected.
- [44:45](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2685s) Local versus cloud models; lost sessions and the "regenerate your prompt" trick; naming AIs.
- [48:25](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2905s) Erwan: not an improved search engine; energy use; the blockchain comparison.
- [50:36](https://www.youtube.com/watch?v=_nSE1CFa_50&t=3036s) Others: AI trained on AI output, fewer Stack Overflow posts, keywords that make models hallucinate, a talk by Sébastien Blanc where two AIs talked to each other.
- [54:14](https://www.youtube.com/watch?v=_nSE1CFa_50&t=3254s) Erwan's worry: blinkered access to information.
- [56:03](https://www.youtube.com/watch?v=_nSE1CFa_50&t=3363s) Wrap-up; Gilles's closing song: "People Love" by Boris Way.

## Notable quotes
> "Rien ne remplace la rencontre d'humain à humain pour parler de ces sujets-là." — Erwan Rougeux, [04:46](https://www.youtube.com/watch?v=_nSE1CFa_50&t=286s)
*"Nothing replaces meeting human to human to talk about these subjects."*

> "Ce n'est pas juste un rendez-vous sympa entre geeks qui vont discuter de JavaScript." — Erwan Rougeux, [05:46](https://www.youtube.com/watch?v=_nSE1CFa_50&t=346s)
*"It's not just a nice meet-up of geeks chatting about JavaScript."*

> "Lui, il est juste là pour te traduire une idée et un besoin en langage informatique, et au final, c'est ce qu'on fait." — Erwan Rougeux, [40:26](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2426s)
*"It's just there to translate an idea and a need into a programming language, and in the end that's what we do."*

> "Là, tous les projets n'ont pas besoin d'IA." — Erwan Rougeux, [49:52](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2992s)
*"Not every project needs AI."*

## Questions this answers
**Q: Why does Erwan Rougeux think public authorities should support tech conferences?**
A: Because they are where professionals share expertise, which matters to the economy; he contrasts France with Côte d'Ivoire, where ministries invest in such events. See [05:17](https://www.youtube.com/watch?v=_nSE1CFa_50&t=317s) and [09:00](https://www.youtube.com/watch?v=_nSE1CFa_50&t=540s).

**Q: Can someone build a real app with ChatGPT without knowing the language?**
A: Erwan built a handpan tablature web app in Rust by describing his need and pasting compiler errors back, in about 150 exchanges over a few nights. It is not perfect, he says, but it exists. See [32:42](https://www.youtube.com/watch?v=_nSE1CFa_50&t=1962s) and [35:58](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2158s).

**Q: How do the hosts use AI for everyday development?**
A: Gilles uses Mistral as a natural-language search engine; Valentin uses it to explain code and GitHub Copilot to speed up tasks he already knows; Erwan pushes it to build whole projects. See [27:14](https://www.youtube.com/watch?v=_nSE1CFa_50&t=1634s), [28:22](https://www.youtube.com/watch?v=_nSE1CFa_50&t=1702s) and [30:10](https://www.youtube.com/watch?v=_nSE1CFa_50&t=1810s).

**Q: Local LLM or ChatGPT: what is the difference in practice?**
A: For Erwan, mainly speed: with 8 GB of GPU, a local answer takes a minute or more, against about 15 seconds for ChatGPT. See [45:06](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2706s).

**Q: How do you keep context when a ChatGPT session is lost?**
A: Erwan asks the model, before leaving, to write a prompt that would restore its parameters, and has automated a daily summary of this kind. See [47:18](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2838s).

**Q: Is using ChatGPT as a search engine a good idea?**
A: Erwan thinks not: it burns GPU and energy for little, while asking it to write a program that then runs on an efficient host makes more sense; on Clever Cloud, he says, you save at least about 30% of the electricity ([49:30](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2970s)). See [48:25](https://www.youtube.com/watch?v=_nSE1CFa_50&t=2905s).

## Mentioned
Clever Cloud, Touraine Tech, Pubstack, Taskfile, JUG Summer Camp, Quentin Adam, Abidjan, Côte d'Ivoire, Small Business Act, CSS, Internet Explorer 6, NASA, Voyager 1, Carl Sagan, Star Trek, Cursor, GitHub Copilot, IntelliJ, GPT, Claude, Gemini, Mistral, Stack Overflow, Llama 3.1, Ollama, ChatGPT, GPT-4o mini, handpan, MusicXML, SVG, MIDI, Rust, Cargo, Rocket, MuseScore, LinkedIn, Reddit, Sébastien Blanc, Boris Way