Jeremy Howard interview at PytorchCon with Anna Tong

2026-02-01 · Source: Jeremy Howard · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Emerging Technologies & Innovation · Depth: Intermediate, long

Summary

Jeremy Howard, co-founder of Fast.ai and Answer.ai, discusses the early days of PyTorch and the development of ULMFiT, which he identifies as the first large language model. He recounts the initial skepticism surrounding PyTorch and language models, contrasting it with Google's TensorFlow and the prevailing belief that language models were not the future. Howard emphasizes the importance of transfer learning and fine-tuning, noting that ULMFiT, developed in a Jupyter Notebook, significantly outperformed existing benchmarks. He also shares his contrarian view on AI agents, arguing that over-reliance on them leads to skill degradation and reduced long-term productivity. Instead, Howard advocates for AI as a tool to enhance human capabilities, detailing Answer.ai's "Solve it" environment, which promotes iterative, human-centric AI collaboration based on principles from George Pólya's "How to Solve It" and Eric Ries's "The Lean Startup." He concludes by stressing the critical role of open-source AI for democratic access and power distribution, citing Meta and NVIDIA as current leaders in open-source model development.

Key takeaway

For machine learning engineers and developers weighing AI integration strategies, avoid over-reliance on AI agents that automate entire tasks. Instead, focus on using AI as a collaborative tool to enhance your skills, learn new concepts, and refine your craft. This approach, exemplified by Answer.ai's "Solve it" environment, promotes sustainable high-quality work and prevents skill atrophy, ensuring your long-term relevance and competence as AI capabilities advance.

Key insights

AI should augment human skills and agency, not replace them, fostering continuous learning and distributed power.

Principles

• Transfer learning is foundational for effective language models.
• Distributing AI power via open source is crucial for democracy.
• Human agency and skill development are paramount in AI interaction.

Method

Answer.ai's "Solve it" environment uses small, iterative steps with AI guidance, focusing on human improvement and skill development rather than full automation, drawing from Pólya's "How to Solve It" and "The Lean Startup" principles.

In practice

• Use AI to improve your craft, not outsource it.
• Prioritize learning and skill growth over short-term AI-driven speed.
• Engage with open-source AI models for broader access.

Topics

• PyTorch
• Large Language Models
• Transfer Learning
• AI Agents
• Open-Source AI

Best for: AI Researcher, Machine Learning Engineer, Entrepreneur

Related on AIssential

• You're Using AI to Learn Faster. That's Exactly Why You're Not Learning — Over-reliance on AI for learning impedes deep understanding and skill development by circumventing necessary cognitive struggle.
• Claude has some conflicts — AI agents are evolving to interact with computers more human-like, driving significant shifts in development and open-source practices.
• bh-AI: So today’s lesson is a little technical so kindly bear with me… Continue reading on Medium » — AI models, while helpful, require human oversight and critical thinking to mitigate hallucinated outputs and ensure effective use.
• [D] Self-Promotion Thread — The community is actively developing diverse open-source tools and frameworks for ML education, research, and practical applications.
• Microsoft’s open-source SkillOpt automatically upgrades AI agent skills without touching model weights — SkillOpt applies deep-learning optimization to text-based agent skills, enabling autonomous, verified performance improvements without altering model weights.
• The Skills That Will Matter When AI Can Do Almost Everything — Human skills that AI cannot replicate will become increasingly valuable as automation expands.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Jeremy Howard.