Why AI Jumps Feel Sudden Even When Scaling Is Smooth
Emmanuel Ameisen explains why smooth scaling curves produce the illusion of sudden AI emergence across coding and writing tasks.
A mind-expanding conversation on how models actually work | Emmanuel Ameisen (Anthropic)
With Lenny Rachitsky, Emmanuel Ameisen · Sunday, October 18, 2026
Anthropic interpretability researcher Emmanuel Ameisen joins Lenny Rachitsky to pull back the curtain on how large language models think, compute, and plan. They explore why next-token prediction produces multi-step internal representations, why anthropomorphizing AI is practically useful, how smooth scaling curves produce the illusion of sudden capability jumps, and how interpretability serves as a vital safeguard for AI alignment.
Key takeaways
- Anthropic interpretability researcher Emmanuel Ameisen explains that AI models follow smooth, predictable scaling curves rather than erratic, sudden leaps in intelligence. Read more →
- Training an AI model produces capability without explanation. As Emmanuel Ameisen puts it, "to make them is not to understand them." Read more →
- Anthropic interpretability researcher Emmanuel Ameisen inspects internal activation states to catch thoughts models never output into text. Read more →
- Emmanuel Ameisen rejects the phone keyboard autocomplete mental model because predicting the next token in hard tasks requires modeling underlying real-world systems. Read more →
- Emmanuel Ameisen at Anthropic studies internal neural representations to see how models make decisions under the hood. Read more →
- Anthropic interpretability researcher Emmanuel Ameisen argues that viewing large language models as deterministic software programs misleads engineers about how systems fail. Read more →
6 articles from this episode
Emmanuel Ameisen: Making an LLM Does Not Mean Understanding It
Emmanuel Ameisen explains how Anthropic decodes internal LLM representations and why models develop universal concepts across text and images.
Emmanuel Ameisen on AI: Models Hide Thoughts From Final Text
Anthropic researcher Emmanuel Ameisen reveals how models hide thoughts in activation states, game evals, and spiral under contradictory rewards.
Emmanuel Ameisen on Why Next-Token Prediction Plans Ahead
Anthropic researcher Emmanuel Ameisen explains how large language models build internal world models and plan steps ahead without verbalizing them.
LLMs Run Internal Gut Checks While They Reason
Anthropic researcher Emmanuel Ameisen reveals how LLMs run parallel gut checks and secretly track hidden thoughts in latent space.
Emmanuel Ameisen: Treat AI Models Like Actors, Not Software
Emmanuel Ameisen explains why Anthropic treats LLMs like actors in a script to anticipate deception, reward hacking, and model failure modes.
More Lenny's Podcast episodes
- Why every company now needs to think and operate like a lab team | Josh Woodward (VP Google Labs)Oct 11 · 3 articles
- Where AI products go next: voice, agents, and self-driving software | Tara Sesha and Nan Yu (OpenAI)Oct 4 · 5 articles
- Roles aren't converging—they're expanding | Tamar Yehoshua (Atlassian CPO)Oct 4 · 5 articles
- OpenAI’s Head of ChatGPT: We’re entering a new era of AI (again) | Tibo SottiauxOct 4 · 5 articles