How are LLMs trained?

LLMs are trained using a process called self-supervised learning on massive datasets of text and code, where the model learns to predict missing words or the next word in a sequence.

Can LLMs think or have consciousness?

No, LLMs are sophisticated pattern-matching systems. They do not possess consciousness, beliefs, or understanding in the human sense; they generate responses based on the data they were trained on.

What are the limitations of LLMs?

LLMs can sometimes generate incorrect information (hallucinate), exhibit biases present in their training data, and may struggle with complex reasoning or tasks requiring real-world interaction.

GlossaryModelsNew to AI

Large language model

Also written: LLM, LLMs, Language Models, LMs, Augmented LLM, LLM-driven, LLM powered, LLM-based reasoning

A type of artificial intelligence trained on vast amounts of text data to understand and generate human-like language.

Explain like I'm new

In plain terms

A large language model, or LLM, is a type of artificial intelligence that has been trained on a massive amount of text. Think of it like a super-powered autocomplete that can understand context and generate coherent sentences, paragraphs, or even longer pieces of writing.

These models learn patterns, grammar, facts, and reasoning abilities from the data they are trained on. This allows them to perform a wide range of language tasks, such as answering questions, summarizing text, translating languages, and writing different kinds of creative content. For example, you can ask an LLM to explain a complex topic like quantum physics in simple terms, and it will generate an explanation based on its training data.

The expert version

Large Language Models (LLMs) are deep learning models, typically based on transformer architectures, trained on massive datasets of text and code. They excel at understanding context, nuance, and generating human-like text across a wide variety of tasks. The scale of their training data and parameter count is a key differentiator, enabling emergent capabilities not present in smaller models. Tradeoffs often involve computational cost for training and inference, potential for factual inaccuracies or biases inherited from training data, and challenges in controlling their output precisely.

Why it matters now

Large language models are central to current AI development, with labs like Anthropic, Meta, and Google exploring their use in everything from code assistance and presentation generation to user-aware agents and ad systems. Recent research focuses on improving LLM capabilities in areas like long-term memory, user context adaptation, and secure data access, highlighting their growing role in practical AI applications.

Where it shows up 148 items, newest first

How each company uses it

Anthropic Claude: Anthropic's recent work shows LLMs being used for code error detection, presentation creation, proactive workspace organization, and as tools for analyzing private data, alongside research into their security as authorization gates.
Google Gemini: Google Gemini is leveraging LLMs, potentially within its developer ecosystem, to enable faster building of AI agents.
Meta Meta AI: Meta AI is scaling its generative ads model, a large language model, to enhance ad performance across its platforms.
xAI Grok: xAI Grok is mentioned as a company involved with large language models.

Often confused with

Artificial intelligence: AI is a broad field encompassing many techniques, while
Machine learning: Machine learning is a method for AI to learn from data; LLMs a
Neural network: LLMs are a type of neural network, but not all neural networks a

Origin

The term 'Large Language Model' emerged as models grew significantly in size (parameters) and the datasets they were trained on expanded dramatically, particularly with the advent of transformer architectures. Source ↗

Try this about 5 minutes

Open ChatGPT, Google Gemini, or Claude and ask it to write a short poem about a robot learning to paint. Pay attention to how it follows the instructions and the creativity of its response. Then, try asking it to explain a simple concept, like photosynthesis, as if you were a five-year-old.

Try it: Playbook entries that use thisPlaybook →

The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.

Case studyUsing LLMs to build tools for iOS location data analysisResearchers used an LLM to build a tool that decoded iOS location data, overcoming LLM analysis limitations. · 60 min
HeuristicIterative refinement for data science agent planningRefine data science agent plans iteratively using a judge and router to improve solutions and handle complexity. · 5 min
HeuristicLLMs are unreliable for direct privacy analysis but good toolmakersUse LLMs to build analysis tools, not for direct forensic or privacy analysis. · 15 min
Eval methodCheck LLM state policy recall for cross-jurisdiction substitutionVerify if an LLM provides another state's policy value when asked about a specific state. · 60 min
Policy noteBuild systems to enforce LLM usage guidelinesImplement technical measures to ensure compliance with LLM usage policies. · 30 min

Questions people ask

How are LLMs trained?: LLMs are trained using a process called self-supervised learning on massive datasets of text and code, where the model learns to predict missing words or the next word in a sequence.
Can LLMs think or have consciousness?: No, LLMs are sophisticated pattern-matching systems. They do not possess consciousness, beliefs, or understanding in the human sense; they generate responses based on the data they were trained on.
What are the limitations of LLMs?: LLMs can sometimes generate incorrect information (hallucinate), exhibit biases present in their training data, and may struggle with complex reasoning or tasks requiring real-world interaction.

First seen on Subvolts Feb 4, 2021. Updated Oct 19, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.

Large language model

Questions people ask

OpenAI (ChatGPT)

Google (Gemini)

Anthropic (Claude)

Microsoft (Copilot)

Meta (Meta AI)

xAI (Grok)

Across the six

Subvolts