MLWhiz Weekly Recsys/ML/GenAI Newsletter # 7 - The week Karpathy Joined Anthropic
Summary
Andrej Karpathy, co-founder of OpenAI and former head of Tesla AI, joined Anthropic on May 19, 2026, to lead a new team focused on accelerating pre-training research using Claude. This move, alongside other senior technical leaders joining Anthropic, signals a significant industry shift. Anthropic also acquired Stainless, a dev tools startup, announced a \$200M partnership with the Gates Foundation, and is reportedly in talks for a \$30-50B raise at a ~\$950B valuation, with a potential IPO by October 2026. Other notable developments include Cerebras's May 15 IPO, which saw its valuation reach \$66B after raising \$5.5B, and the release of Alibaba's Qwen 3.7 model. Research highlights include Snap's SID-MLP showing MLPs can outperform Transformers for generative recommendation, Microsoft's HyDRA saving 54.1% in GitHub Copilot costs, and Meta's MARS proposing an agentic RecSys with hierarchical memory. Pinterest also shared a guide for testing AI agent skills in production.
Key takeaway
For AI Scientists and Machine Learning Engineers evaluating strategic career moves or platform investments, Andrej Karpathy's decision to join Anthropic, coupled with their rapid funding and acquisitions, suggests a strong bet on Anthropic's recursive self-improvement approach to LLM development. You should closely monitor Anthropic's technical advancements and consider how their "model making model better" strategy could influence your own research or product roadmaps. Additionally, explore cost-saving techniques like model routing and efficient generative recommendation architectures.
Key insights
Top AI talent and significant capital are rapidly consolidating around Anthropic, signaling a strategic industry shift.
Principles
- AI chip demand remains structurally undersupplied.
- Recursive self-improvement drives LLM development.
- Cost optimization is a key LLM deployment focus.
Method
A ModernBERT encoder routes queries across specialized LLMs based on reasoning, code generation, debugging, and tool-use dimensions for cost-efficient deployment.
In practice
- Consider MLPs for generative recommendation tasks.
- Implement systematic testing for AI agents in production.
- Explore agentic RecSys with hierarchical memory.
Topics
- Anthropic
- Andrej Karpathy
- Large Language Models
- AI Chips
- Recommendation Systems
- AI Agent Testing
- Model Routing
Best for: Investor, Research Scientist, AI Scientist, Machine Learning Engineer, Director of AI/ML
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by MLWhiz: Recs|ML|GenAI.