Jerome Pesenti, VP of AI at Facebook, on using AI for content moderation, limitations of deep learning and current AI, commercializing AI research, and more
Jerome Pesenti is encouraged by progress in artificial intelligence, but sees the limits of the current approach to deep learning. Tweets: @went1955 and @wired Tweets: Robert Went / @went1955 : Facebook's Head of AI Says the Field Will Soon ‘Hit the Wall’. Jerome Pesenti is encouraged by progress in artificial intelligence, but sees the limits of the current approach to deep learning https://www.wired.com/... @wired : Despite some notable AI flops, Facebook continues to use AI to build new features and products, from Instagram filters to augmented reality apps. @willknight talked to Jerome Pesenti, head of AI at Facebook, about AI's future at the company: https://www.wired.com/...
Context & Ripple Effects
The interview lands mid-arc for Facebook's AI organization: Yann LeCun built the research lab around deep learning starting in 2015, then handed day-to-day leadership to Jerome Pesenti in 2018 while staying on as chief AI scientist. By this point AI is embedded across the company — Instagram filters, AR apps, and increasingly the moderation pipeline, where Mike Schroepfer had already been touting progress on detecting violating images, video, and multilingual text months earlier.
What makes Pesenti's position notable is the tension: he runs an operation betting heavily on deep learning in production while publicly warning the field will soon hit the wall of the current approach. His answer — commercialize the research — frames FAIR less as a science lab and more as a product engine.
First-order effects
- Facebook's moderation systems, which Schroepfer's team has been scaling across image, video, and text, now carry their own leadership's caveat that the underlying technique has hard limits — raising the stakes on human review capacity where models fall short.
- Pesenti's commercialization push puts FAIR researchers on a path toward shipping products rather than publishing alone, changing what the lab optimizes for under his tenure.
Second-order effects
- If deep learning plateaus as Pesenti suggests, rivals that have tied their product roadmaps to the same technique family face the same ceiling, making research breadth — not model size — the differentiator among big-lab players like Facebook.
- Moderation tooling vendors and platform trust-and-safety teams inherit the gap: platforms leaning on AI enforcement will need hybrid human-machine workflows wherever the wall bites first.
Third-order effects
- The pattern points toward AI labs consolidating into industrialized research arms attached to distribution — a shift already visible in how Facebook folded applied machine learning into core strategy — with pure-research independence becoming harder to sustain.
- If the wall proves real, the next competitive cycle belongs to whoever funds post-deep-learning approaches before their production systems saturate, turning fundamental research back into a strategic necessity rather than a cost center.
The trend: Big-platform AI labs are shifting from open-ended deep learning research toward commercialized, product-bound operations even as their own leaders flag the technique's approaching limits.