Arthur Mensch (Mistral AI) and John Collison (Stripe) fireside chat | Stripe AI Day—Paris
Arthur MenschSpeaker- John CollisonHost
Key ideas
Mensch argues that open weights are both Mistral's business edge and the better route to safety, cost control and production use for businesses.
From a machine transcript of the audio. Quotes link to the moment they are said.
- 1
Compete with an open-core model
Mensch describes Mistral's main difference from the large US providers as releasing model weights and aiming for the best open models, with proprietary models and hosting alongside. He says Mistral was started because he sees an oligopoly forming in the US and thinks Europe would be too late if it did not act soon.
- 2
Expect open models to win in production
Mensch says most businesses are still at proof of concept and not very cost sensitive, but that production deployments will need control over cost, infrastructure and the model itself. He points to browsers, operating systems and Kubernetes as earlier markets that ended up open source.
- 3
Safety needs control, not a black box
Mensch argues a strong model must be able to process harmful content in order to recognize it, and that downstream users can build moderation and guardrails better with access to the weights. He says he thinks there is a strong misunderstanding about safety, and that the real goal is models that can be controlled.
- 4
Regulate documentation, not hard thresholds
Mensch's preferred rule is auditable models with documentation proportional to the user base. He opposes rules that regulate a model once it passes a fixed compute or benchmark number, saying they do not fit a technology that keeps growing.
- 5
Build AI intuition through compute and time
For engineers moving into AI, Mensch recommends starting with deep learning basics and training a small model before heading into production work. He says intuition about what works is the biggest value in good AI engineers, and that it takes compute, experiments and patience.
In Arthur’s words
“Currently we are seeing on the US side some kind of oligopoly shaping up.”
Arthur Mensch·Arthur Mensch (Mistral AI) and John Collison (Stripe) fireside chat | Stripe AI Day—Paris·Watch at 28:01(opens the original in a new tab)“When we move into production, when those enterprises decide to deploy to their end users some generative AI solutions, they will need to control the cost. They will need to control the infrastructure.”
Arthur Mensch·Arthur Mensch (Mistral AI) and John Collison (Stripe) fireside chat | Stripe AI Day—Paris·Watch at 22:21(opens the original in a new tab)“What we really want from models is for them to be controllable. And you want the downstream users, the users that create some applications, to take this model and control them into an end user application.”
Arthur Mensch·Arthur Mensch (Mistral AI) and John Collison (Stripe) fireside chat | Stripe AI Day—Paris·Watch at 5:41(opens the original in a new tab)About this entry
- Published
- Oct 6, 2023
- Channel
- Stripe
- Type
- Video
- Length
- 38:50
- Language
- English
- Added
- Nov 2, 2026
- Source last checked
- Nov 2, 2026
What’s available here
Original source
Linked. Opens on YouTube.
Player on this page
Available. YouTube’s own player, loaded only when you press play.
Full text or transcript
Not reviewed. No license to show a text version here.
Our summary
Draft. Written by us with AI from a machine transcript of the video’s audio (Whisper) and checked against it. Machine transcripts can mishear; each quote links to the moment so you can hear it. Not yet rights-reviewed.
Ask questions with AI
Not reviewed. Off until there is licensed text to work from.