Back to Reading List
[Scaling]·PAP-MJCGEH·March 17, 2026

Emergent Abilities of Large Language Models

Jason Wei, Yi Tay, Rishi Bommasani et al.

4 min readScalingReasoning

Core Insight

Larger language models develop unexpected skills, challenging our predictions and scaling strategies.

Origin Story

arXiv preprint, May 2022Google ResearchJason Wei, Yi Tay et al.

The Room

Inside Google Research, a group of curious minds huddled together in a conference room. They were captivated by a lingering question: What uncharted territories lay beyond the known scaling laws of language models? The room buzzed with a mixture of excitement and skepticism as they grappled with the unknown, wondering if larger models could surprise them in unexpected ways.

The Bet

The team decided to push the limits further than ever, betting that by scaling models massively, they might stumble upon unforeseen abilities. It was a risky move, as the hypothesis sounded almost too optimistic. Doubts lingered, and there was a moment when submitting the paper felt like leaping into the dark, unsure if their hunch would hold any water.

The Blast Radius

Without this leap of faith, the landscape of AI would be starkly different. Models like PaLM and Claude might not exist, and the narrative around emergent properties would be less vibrant. Jason Wei and Yi Tay have continued to explore these themes, influencing new research directions and inspiring the next wave of AI breakthroughs.

PaLMClaudeGemini

Knowledge Prerequisites

git blame for knowledge

To fully understand Emergent Abilities of Large Language Models, trace this dependency chain first. Papers in our library are linked — click to read them.

DIRECT PREREQIN LIBRARY
Attention Is All You Need

Understanding the attention mechanism is fundamental for grasping how large language models work since they rely heavily on transformer architectures introduced in this paper.

Attention mechanismTransformer architectureSelf-attention
DIRECT PREREQIN LIBRARY
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

This paper introduces Bidirectional Encoder Representations from Transformers, which is a foundational large language model that shows how pre-training on vast textual data can improve language understanding tasks.

Bidirectional transformersPre-trainingMasked language models
DIRECT PREREQIN LIBRARY
Training language models to follow instructions with human feedback

Understanding instruction following through human feedback is crucial for realizing how large language models can be fine-tuned to improve task performance based on human-provided feedback.

Fine-tuningHuman feedbackInstruction-following
DIRECT PREREQIN LIBRARY
Toolformer: Language Models Can Teach Themselves to Use Tools

This paper explores how large language models can utilize external tools to enhance their capabilities, a concept that is potentially linked to the emergent abilities described.

Tool use in language modelsSelf-improvementExternal API integration
DIRECT PREREQIN LIBRARY
Tree of Thoughts: Deliberate Problem Solving with Large Language Models

Understanding structured thinking processes in language models will provide insights into how these models develop emergent problem-solving abilities.

Tree structuresDeliberate reasoningProblem-solving strategies

YOU ARE HERE

Emergent Abilities of Large Language Models

In Plain English

The paper reveals that exhibit absent in smaller ones, defying performance predictions. These findings suggest that mere scaling introduces novel capabilities that smaller models can't achieve.

Explained Through an Analogy

Imagine a plant that not only grows taller with water but blooms new, unforeseen flowers. Scaling models similarly reveals hidden algorithms.

Go deeper for $6/mo

Everything a PM needs to turn this paper into a competitive edge — in under 10 minutes.

  • 2-page deep-dive article
  • Highlighted key passages
  • Expert-mode reading layer
  • PM Action Plan — 3 moves
  • Use cases for your product
  • Meeting talking points
  • Interactive paper simulator
  • Test Your Edge quiz

Already subscribed?

Log in

How grounded is this content?

Metrics are computed from available source text only — abstract, summary, and impact fields ingested into this system. Full paper PDF is not ingested; numerical claims that originate from within the paper body will not appear in these scores.

Source Richness88%

7 of 8 content fields populated. More fields = better-grounded generation.

Source Depth~225 words

Total source text analyzed by the model. Includes extended deep-dive summary — high confidence.

Methodology: Number grounding uses regex digit extraction against source text. Quote traceability uses token set intersection on content words stripped of stop-words. Neither metric validates semantic correctness or factual accuracy against the original paper. For full verification, cross-reference with the original paper via the arXiv link above.