Some Quick Reflections on DeAI at ICML 2025: YoY Growth in Presence, Quality of Decentralized AI Contributions It was highly encouraging to see a step up in presence from top DeAI teams – not just as attendees but as core participants presenting novel papers + research From @ritualnet and @exolabs to @PluralisHQ, @gensynai, @PrimeIntellect, @bagelopenai, @akashnet_ and others – the quality of contributions to the field is increasingly outstanding, original, and bringing legitimacy to crypto as a domain where fundamental innovation in AI can credibly take place We’re Still (Very) Early That said, it remains the case that for most of the conventional AI field, ‘decentralization’ isn’t really a relevant part of the conversation Two Obvious Reasons For This: 1) DeAI is largely still in its R&D era – more research than production-grade reality 2) DeAI is preemptive – from training to deployment, the constraints that’ll drive relevance + adoption of the emerging decentralized stack do not (yet) outweigh the practical benefits that the traditional stack offers Early is Okay, BUT: Ideology can’t replace technical or economic parity. To really have an impact, it’s critical that teams look beyond the crypto silo to understand what builders + users are actually optimizing for. Naturally, it'll be (user / product) dependent and vary across a spectrum of reliability, performance, cost, compliance/security, privacy, new economic models and more But the ability to do this effectively will determine whether the decentralized AI stack can be truly transformative across both worlds, or if it remains a performative exercise in launching tokens