Me, Myself, and AI From Rabbit Holes to Recommendations: Reddit’s Vishal Gupta
20 snips
Nov 11, 2025 Vishal Gupta, an engineering manager for machine learning at Reddit, shares insights from his extensive background at DeepMind and Google. He dives into how Reddit leverages sophisticated AI to enhance user experience and ad relevance. Vishal explains the evolution of recommendation systems from basic methods to complex deep learning techniques, stressing the need for balance between user engagement and advertiser goals. He champions the value of authentic human conversation amidst the rise of AI content, making a compelling case for collaboration between human and machine insights.
AI Snips
Chapters
Transcript
Episode notes
Recommendation Systems Power Discovery
- Recommendation systems capture short-term and long-term user interests to suggest valuable content.
- They also support search by satisfying explicit user queries with relevant items.
Three Waves Of Recommenders
- Recommender tech evolved from collaborative filtering to deep learning and now LLM/multimodal representations.
- Modern systems use rich text and visual embeddings for hyper-personalized recommendations.
Avoiding The Rabbit Hole
- Optimizing purely for past behavior risks driving users into 'rabbit holes' of similar content.
- Systems must balance exploration and exploitation, sometimes using RL or heuristics.