Project summary

Paris has world-class AI safety institutions and is still lacking a front door, which I am building through Nouvelle Machine : a weekly bilingual dinner, and a Saturday build club (called Banc d'Essai) that intends to turn curious newcomers into people doing real AI-safety work, then route them to organizations that fit their interest and highest possible impact.

What are this project's goals? How will you achieve them?

The goal is to convert latent French talent into people who have touched AI-safety work and adeptly been routed onward to the institutions that can use them. Concretely, by the end of six months, I am aiming for >25 regulars, >15 first projects shipped, and >8 confirmed onward routings into organizations like CeSIA, BlueDot, ML4Good or a governance track. I have been tracking these numbers from day one, and they will be reported publicly at 3 and 6 months.

Nouvelle Machine has two linked formats, that are both live since the September rentrée.

Dîner Machine is a weekly bilingual (FR/EN) tech dinner. It's the top of the funnel: it gathers open, capable people (engineers, students, career-changers) who don't yet identify with AI safety. Each dinner has a theme/rotating edition: a papers night, a game-theory night, a predictions night, etc. I believe bilingual-by-default can make a real difference, in a scene that otherwise splits into francophone-institutional and anglophone-expat with almost no bridge.

Banc d'Essai is the Saturday build club, with the first session planned for October 17th. The format is: a short accessible demo to open (run a small eval, read a transcript of a model scheming), then cowork blocks where each person starts a first small project: an eval, a French translation of a key resource (I will especially be focusing on those at the start, for further outreach), a short governance brief, a rebuttal post. I keep a project menu to prevent the failure mode of people not knowing which project to start and leaving without having gotten first-hand experience.

My thinking on routing, especially, has moved since the spring. I used to pitch one of the outcomes of this as producing (future) junior technical researchers, and I don't believe that framing anymore, as I now believe governance is much higher-impact than technical safety. Doing a small project, as I see it, is mostly the fastest way to actually understand the problem and to feel it's yours to work on. Furthermore, I think AI safety needs to become a broad cause the way climate is one (as I have written in my article): it needs technical people, but also governance, policy, communication and an informed public around all of them. So governance and comms are explicit exit paths in how I route people, and in fact, I will likely prioritize those at the start (depending on the profiles).

I focus on the entry level routing, which makes me a feeder, and not a rival, of the existing institutions; many of which I am on good terms with (CeSIA, ML4Good, and the upcoming impact hub in Paris). I intend to use my network to help identify competent people and put them in contact with the institutions that best fit their areas of interest, and their highest leverage for impact.

If it works it becomes a replicable kit, which I am interested in moving towards eventually in Europe, notably because of how neglected the middle powers have been in the AI Safety conversation thus far.

How will this funding be used?

I'm asking for €17,400 to €23,100 (roughly $20,000 to $27,000) for six months, depending on how much of the operations budget other funders cover. About 75% is my salary (6 × €2,900/month, self-employed, includes ~22% French cotisations), which is around 70% of my Paris counterfactual salary. The rest is operations, to cover the build club (food, room, compute for demos and attendee projects), a Demo Day where people can present what they built, and an ~8% buffer. Attendees pay for their own dinner at Dîner Machine, so there's no food line for the weekly dinners. The minimal version is salary-only, if the operations lines are covered by Essaim or a small BlueDot grant.

Who is on your team? What's your track record on similar projects?

I, single handedly, have built this kind of community from zero before. In Toronto I ran a weekly ritual that grew into a community of around 100 people, plus an events-aggregation platform and one-off formats (demo nights, a cohesion summit, teaching at Weave). I launched AI-safety meetups from zero in under two weeks each in Kraków and Warsaw. I attended 110+ events in a year with systematic notes, so I know empirically what makes a room work. I'm a native FR/EN bilingual, resident in Paris, in contact with CeSIA and around the Pause IA France, EA France, and ACX/LW crowds. I build my own tooling, including a Paris event scraper and classifier, as well as my own CRMs and outreach automation. On the same thesis I am launching a French language AI Safety Digest newsletter, have started short-form French AI-safety video, and wrote a manifesto article on making AI safety a mainstream cause through education (thelastsocialite.substack.com, French version at nouvellemachine.com).

What are the most likely causes and outcomes if this project fails?

The most likely failure, and core risk, given the thesis, is that people come but don't convert: the dinner gathers a nice crowd but few of them ship a first project. I'm addressing it with a small-granularity project menu (2 to 3 hours) and hands-on help in the room, but it's the hypothesis the first three months will actually test, and I'll report the real numbers either way. The second risk is Paris retention, since the city is saturated with events and the crowd is volatile; I recruit in person at existing events and launched at the rentrée, the best window of the year. The third is the single-failure point: the whole project so far depends on one person, me. If it fails, the outcome is mild and cheap: the dinner shrinks to hobby scale or merges into an existing event, fewer francophones get routed in, and I publish an honest write-up of what didn't work so the next person doesn't repeat it.

How much money have you raised in the last 12 months, and from where?

None so far. Essaim, the francophone EA funding circle, has decided to fund this project; the exact amount is being finalized and expected within days. Nothing else has been raised in the last 12 months. I'm also applying to the LTFF and the Astralis/Macroscopic Middle Powers RFP, and I'll disclose the Essaim figure and reduce overlapping asks as soon as it's confirmed.